Aside from the MAGA base, everyone has polling PTSD. Leftists, liberals, centrists, and Never Trumpers do not trust what their eyes and ears are telling them: that the Trump coalition is damaged, weak, and wobbling. The midterms really might usher in the beginning of the end of MAGA.

Part of the public’s skepticism is that the upper echelons of the tech industry are doing their best to keep Republicans financially competitive. Tech oligarchs, as well as many AI and crypto super PACs, are spending big to elect candidates (especially conservatives) whom they believe will maintain a light-touch regulatory approach.

Tech oligarchs are aligning with Trump because of their own business interests, certainly, but they’ve been radicalized, too. Elon Musk is chief among them. He’s given more than $400 million to swing these midterms back to Republicans—an unseemly number with no precedent.

Still, the signals are impossible to ignore. Trump is unusually unpopular for a president at this stage of his term. Democratic campaign committees have far outraised their Republican counterparts, which points to an enthusiasm gap among rank-and-file contributors. Republicans, meanwhile, are dwarfing Democrats in super PAC funding. Campaign committees have capped donation limits, super PACs don’t. The most influential tech donors are trying to buy their preferred election results via those super PACs. This time, I don’t think they’re succeeding.

As a politics nerd who’s been studying early turnout reports, polling, and campaign donations, I decided to put together a list of 11 crucial races where the tech industry has played an outsized role in funding candidates (or informing their messaging). Why 11? Because I reject nice, round numbers. To put my money where my mouth is, I’ve issued predictions for each race.

Shoutout to two resources that I found especially useful: this CNBC article about AI political groups, and this database from Molly White’s Citation Needed.

Arizona’s 1st Congressional District

The candidates: Jay Feely (R), Amish Shah (D)

Big Tech’s influence on the race: At least $616,000 to support Feely via the American Mission super PAC, which gets behind “Republican candidates dedicated to advancing AI innovation.” In other words: AI regulation not good.

The stakes: This is a true swing district in a swing state. The outcome has a pretty decent shot at reflecting the broader electorate in other must-win House races. The candidates are not very remarkable. Shah is a moderate Democrat. Feely is MAGA-adjacent and has spent the last few months suggesting to voters that he isn’t a Trump loyalist.

Shah said constituents in the district flatly do not want any data centers to be built there. Feely has touted the economic benefits of data centers and avoided giving a firm stance on whether they should be constructed in Arizona’s first congressional district.

Hard Reset’s stone-cold lock: Shah wins because independents go for him in droves. “Former professional athlete” only takes you so far when you were a kicker. Huge L for data centers.

California’s 11th Congressional District

The candidates: Scott Wiener (D), Connie Chan (D)

Big Tech’s influence on the race: A Wiener super PAC recently received $3 million from Public First Action, an advocacy group that claims to be nonpartisan and in favor of AI regulations. Public First Action is deep-pocketed because of $40 million in donations from Anthropic, though Anthropic swears its donations are not supposed to be funneled to any one candidate. I have no evidence to the contrary, though I am skeptical of those nebulous guardrails. Chan derided the donation; Wiener had denied any involvement from Anthropic. Separately, the Wiener campaign is reportedly coordinating extremely closely with a different super PAC that favors his candidacy.

Given that this is San Francisco, there are also lots of “maxed out” donations ($3,500) from wealthy tech workers and executives. Here’s a filtered FEC link if you’d like to peruse. (Click the “raw” button on the far left if you want to see more recent donations that aren’t yet itemized by the FEC.)

The stakes: Nancy Pelosi’s legacy is at stake! She endorsed Chan, who appears to be the underdog going into Tuesday. Chan wants a federal moratorium on data centers and has hammered Wiener for not being a trustworthy source on the issue. Wiener has a legitimate history trying to regulate AI in the California State Legislature, though his bolder bill got squashed by Governor Gavin Newsom. My Hard Reset colleague JJ Lansing has a thorough write-up of how each candidate says they’ll legislate on AI.

Hard Reset’s stone-cold lock: Wiener wins, I think. My disclaimer is that I am not following Bay Area politics as closely as I did when I lived there. My sense from afar is that Wiener has consolidated the traditional Democratic base in San Francisco—especially its highest-earners. Mission Local’s analysis of the race noted that, despite Nancy Pelosi’s endorsement of Chan, strong supporters of Pelosi are behind Wiener, 54% to 46%. That’s pretty convincing evidence to me, though the same analysis raises strong points that Chan’s support among monolingual Chinese voters is likely being underrepresented in polling.

California’s Prop. 40

The ballot measure: A 5% tax on California residents with net worths over $1 billion.

Big Tech’s influence on the race: About as sizable as you can get. According to Politico, more than $100 million from Google cofounder Sergey Brin; more than $40 million from former Google CEO Eric Schmidt; more than $33 million from Stripe cofounder Patrick Collison; more than $17 million from Ripple cofounder Chris Larsen; more than $12 million from venture capitalist Vinod Khosla; more than $7 million from venture capitalist Michael Moritz; more than $3 million from DoorDash CEO Tony Xu. Every day, the anti-wealth tax super PAC Building a Better California is burning through millions of dollars.

The stakes: If Prop. 40 passes, and billionaires are eventually compelled to pay up (they will surely challenge the measure in court), then wealth tax discussions will extend well beyond California. A federal wealth tax will be a consistent position among Democratic Party candidates in the 2028 presidential primaries. Also: more healthcare funding in the state of California!

If Prop. 40 fails, there will be lots of finger-pointing. Billionaires will continue hoarding billions. We’ll learn who actually departed the state this year, and who faked it. A few of the Democratic candidates for president will push for a federal wealth tax, but it will sadly lose some salience.

Hard Reset’s stone-cold lock: Here’s what I wrote last weekend, which I feel even more confident about after reading recent polling.

The wealth tax ultimately will not be enacted. If Prop. 40 passes, it can still be undone by Prop. 41 and Prop. 42, which are basically poison-pill measures that billionaires snuck onto the ballot. In the event that either proposition accumulates more votes than Prop. 40, the wealth tax is invalidated. That gives California billionaires three different outs. I do not like those odds, especially coupled with a lack of enthusiasm for candidates at the top of the ticket. (Xavier Becerra is not going to be among the Democrats who outperform expectations.)

Florida’s 25th Congressional District

The candidates: Scott Singer (R), Rep. Jared Moskowitz (D)

Big Tech’s influence on the race: Moskowitz has benefited from at least $310,000 in support from Think Big, a super PAC that claims to back Democrats who are “advancing AI innovation.” Think Big is reportedly an affiliate of Leading the Future, the anti-AI regulation super PAC that OpenAI co-founder Greg Brockman recently (publicly) distanced himself from.

Since the primaries, Singer has been bolstered by Defending Our Values, a super PAC that claims to support conservative candidates “who will stand up to the tech industry” and are “committed to AI policies that protect our children from online harm.” Defending Our Values, an affiliate of Public First Action (the group that gave to Wiener), has reportedly contributed more than $1.5 million to benefit Singer.

This is the most expensive House race in the country: it could eclipse $50 million in total spending by Nov. 3.

The stakes: Moskowitz is among a handful of legitimately conservative Democrats in the House. He is further to the right than many Republicans (including Singer) on how to regulate AI. He’s also arguably the most hawkish, pro-Israel Democrat in the House. And he was the only Democrat to join the DOGE caucus, which is now dead.

Republicans redrew this seat as part of their naked gerrymandering scheme. The fact that it’s still a tossup points to the GOP’s fragile standing.

Hard Reset’s stone-cold lock: Moskowitz wins. Hard to say anything definitive about a local-ish Florida race, but early returns strongly favor a Democratic Party overperformance in the state. Especially for a right-leaning incumbent.

If you’ll allow me one paragraph to go Big Brain mode: A Moskowitz victory is bad for Democrats. He is a thorn in the side of the Democratic caucus. “I go after my party all the time and fight within,” is how he sold his credentials to Fox News in August. House Minority Leader Hakeem Jeffries just inexplicably appointed Moskowitz as the ranking member on the Middle East and North Africa Subcommittee. That decision was panned by the left, and for good reason: Moskowitz is completely out of step with the Democratic base on Israel and the Iran War, which he initially cheerleaded.

This is not to say Singer would be any better on these issues—he certainly would not. But he isn’t a rubberstamp for the AI industry, and he likely wouldn’t be chairing powerful subcommittees. Democrats are going to win the House regardless—this is a race where it’d be to their benefit if Moskowitz went down.

Maine Senate

The candidates: Sen. Susan Collins (R), Troy Jackson (D)

Big Tech’s influence on the race: Musk’s super PAC has reportedly disbursed more than $170,000 to a group that produces mailers on behalf of Collins. The SpaceX PAC gave $50,000 to a Collins super PAC and $55,000 to another Collins PAC. In 2025, Palantir CEO Alex Karp donated $100,000 to the main Collins super PAC. Coinbase CEO Brian Armstrong also gave $12,000 to a Collins PAC. Dark-money groups whose donors will probably never be released have reportedly amassed more than $125 million to help Collins get over the finish line.

The stakes: Collins remaining in office for another six years significantly dents Democrats’ hopes of amassing a Senate majority that can actually pass legislation after 2028. Keep in mind: Democrats are always going to have right-leaning defectors lurking among their ranks. They can’t afford to navigate those lurkers and Collins, one of the only Republican Senators in a blue-ish seat. If Jackson loses, the left is going to eat a bunch of shit—both for elevating him, and for sticking with Graham Platner for too long.

Hard Reset’s stone-cold lock: Jackson wins, but it’s ugly and very close. Was always going to end this way with Susan Collins: she does not go down easily. Pine Tree Results, the top Collins super PAC, will release a post-election FEC filing showing that the tech industry panicked and doled out many millions of dollars to try and save Collins during the last few weeks of October.

Michigan Senate

The candidates: Rep. Mike Rogers (R), Abdul El-Sayed (D)

Big Tech’s influence on the race: WhatsApp co-founder Jan Koum contributed $265,800 to a super PAC that supports Rogers. Michelle Valentine, a tech investor and the wife of Anthropic co-founder Tom Brown, gave $25,000 to a different Rogers PAC. (Brown himself gave $3,500, I previously reported.) And Rogers stands to potentially benefit from his own tech investments, especially if there’s a data center boon, The Lever reports.

The stakes: Oh God, where to even begin? Control of the Senate, for one; there are almost zero paths for the Democrats if El-Sayed falters. El-Sayed is a litmus test for other swing-states. Are these states willing to elect a Muslim to statewide office? Are they willing to elect a left-leaning candidate? Not just that: a left-leaning candidate who’s willing to call the genocide in Gaza a genocide?

Of course, those are the stakes stripped of vital context. El-Sayed has been subjected to Islamophobic attacks and smear attempts by Rogers and Rep. Haley Stevens, the latter of whom was El-Sayed’s primary opponent. Chances are, even if El-Sayed wins, he’ll perform slightly worse than Democratic gubernatorial candidate Jocelyn Benson. Centrist Democrats will frame this as a sign that left-leaning candidates are weaker than moderates.

On the tech front: El-Sayed has proposed an ambitious regulatory framework for the AI industry; Rogers has not. Both candidates claim to favor a data center moratorium in Michigan.

Hard Reset’s stone-cold lock: El-Sayed wins by a semi-comfortable margin—more than what is being represented in recent polls. In the coming weeks and months, Beltway pundits and reporters will slowly wrap their heads around how El-Sayed ended up securing an impressive proportion of independent and right-leaning voters.

Montana’s 1st Congressional District

The candidates: Aaron Flint (R), Sam Forstag (D)

Big Tech’s influence on the race: The American Mission super PAC—which also supports Feely’s congressional run—has reportedly contributed at least $590,000 to Flint.

The stakes: Forstag’s victory would be an affirmation of his campaign strategy, which includes running ads that refer to Flint as “AI Aaron.” Forstag has called for much more AI regulation—and called out AI super PAC donations to Flint.

Hard Reset’s stone-cold lock: Forstag wins this one relatively comfortably. I am bullish on the 32-year-old firefighter who has strong labor credentials and a solid populist agenda. Flint’s soft-touch approach to data centers (he’s generally in favor of them) and his saber-rattling about AI (we have to beat China, blah blah blah) will come back to bite him.

Ohio Senate

The candidates: Sen. Jon Husted (R), Sherrod Brown (D)

Big Tech’s influence on the race: Fairshake, a crypto super PAC with significant Big Tech backing, has reportedly spent at least $55 million to keep Brown—a vocal crypto critic—from returning to the Senate. The Winklevoss twins, literal crypto bros, have thrown in for Husted. So have Marc Andreessen and Ben Horowitz.

The stakes: Trump’s popularity in Ohio, plus tens of millions of dollars from Fairshake, were enough to boot Brown in 2024. Were he to come back and win two years later, it would be a pretty stunning repudiation of MAGA and the crypto industry. Brown has focused many of his attacks on Husted’s wishy-washy data center positions, but it’s not clear yet whether that tactic is swaying voters.

Hard Reset’s stone-cold lock: Husted’s allegedly perfect butt will not save him. Brown wins by less than Acton—I wouldn’t be surprised if it’s very close—but he gets the job done, partially because a not-insignificant portion of the electorate seems to believe he’s the incumbent.

Ohio Governor

The candidates: Vivek Ramaswamy (R), Amy Acton (D)

Big Tech’s influence on the race: Musk emptied one of his pockets and sprinkled $5 million to V-PAC, the Ramaswamy super PAC. TikTok investor Jeff Yass gave $10 million to V-PAC. So did Ratmir Timashev, a tech entrepreneur who aspires to make Columbus, Ohio, “the next Silicon Valley.” I wrote more about Timashev last week.

The stakes: Acton recently said that “decency is on the ballot,” which is the sort of “West Wing”-style sentiment that hasn’t served Democrats well during the MAGA era. But this is a rare exception where she is absolutely correct. Ramaswamy was already one of the worst GOP candidates of this election cycle, even before his last-minute campaign ad that featured Acton’s estranged and unwell mother. Now he’s in the running for the worst GOP candidate of this election cycle. If he somehow wins anyway, it means Ohio is out of reach for Democrats for the foreseeable future.

Hard Reset’s stone-cold lock: Acton wins comfortably, Ramaswamy tries to call her to concede, she doesn’t answer the phone. Good for her. Ramaswamy’s political career is over. His disgusting closing message—endorsed by Ohio’s own JD Vance—causes permanent damage to Vance’s standing in his home state.

South Carolina Senate

The candidates: Sen. Darline Graham Nordone (R), Annie Andrews (D)

Big Tech’s influence on the race: Building the Future, a new, pro-data centers super PAC that has yet to disclose its donors, is reportedly spending at least $1.5 million on ads for Graham Nordone. She’s gotten individual donations from Andreessen and Horowitz. Another PAC of hers has received donations from venture capitalist Asha Jadeja Motwani and the Winklevoss twins.

The stakes: If Andrews wins, it means MAGA is functionally over, even with two years left in Trump’s term. It would also be an inspiring repudiation of the late Lindsey Graham’s warmongering legacy. And a repudiation of nepotism! And data centers! Andrews has referred to Graham Nordone as “Data Center Darline.”

Hard Reset’s stone-cold lock: Gotta take the hopium mask off here. Graham Nordone wins. She is obviously unqualified, and this result is an embarrassment. But two important reality checks: Trump carried the state by nearly 18 points in 2024. Even a “blue tsunami” scenario nationwide, coupled with a notable turnout increase among Black South Carolinians, coupled with Graham Nordone not knowing her left from her right—let alone where Taiwan is on a map—might not be enough to overcome that sort of deficit. This is the sort of race where tech’s late influx of cash for Graham Nordone can make the difference.

Also, I’m sorry, but I watched Andrews debate with Graham Nordone and came away deeply unimpressed. Aside from a few funny barbs directed at Graham Nordone (who’s not exactly a difficult target to ridicule), Andrews’ delivery was abysmal. She’s running scared—she sounds like all of the other losing Democrats from the last decade whose platforms were “I am basically a Republican, I swear.” At one point during the debate, she said, “To set the record straight: I don’t support Medicare for All.”

I don’t expect a South Carolina Senate candidate to run as a Bernie Sanders disciple, but you have to have some convictions and some sort of affirmative message. Not the case in this race. After Trump wrote transphobic slurs about Andrews on Truth Social, she responded by cutting an ad where she clarified that she doesn’t “support biological men playing women’s sports” and that she doesn’t support (and has never performed) gender-affirming surgeries for minors. Weak and undignified. If Andrews pulls out an upset, it’ll be a hilarious defeat for Trump and the Graham Dynasty. Just be prepared for Andrews to become one of the worst Senators in the Democratic caucus.

Texas Senate

The candidates: Ken Paxton (R), James Talarico (D)

Big Tech’s influence on the race: Well, Elon Musk might rig this one by himself. Out of the more than $400 million he’s spent on these midterms, roughly $190 million has gone towards rescuing Paxton, according to The Texas Tribune.

Talarico-associated super PACs have received huge donations that would normally be considered impressive, or concerning, depending on your perspective on these things. LinkedIn cofounder Reid Hoffman gave $5 million to the main Talarico super PAC; Netflix CEO Reed Hastings gave $1 million.

The stakes: If Paxton wins, the GOP will breathe a sigh of relief. Democrats’ odds of taking the Senate drop precipitously. Musk and other tech oligarchs prove that they can rig an election with almost half-a-billion dollars in donations—an ominous development heading into 2028. There is no telling how Paxton will repay Musk.

On the flip side, if Talarico beats Paxton, you can assume with near-certainty that the Democrats took the Senate. There’s a less-discussed angle here: Paxton is a uniquely bad candidate, yes, but a Blue Texas (Blexis) almost certainly unlocks an entirely new electoral map for Democrats. Even the threat of Texas being competitive in 2028 means Republicans have to expend more money and resources to ensure they lock up the state’s 40 electoral votes. That helps Democrats elsewhere.

Also: a Talarico victory means Musk wasted a mind-numbing amount of money. I doubt that outcome would compel Musk to back away from conservative candidates, but it might cause him to revisit his stupid idea to start another political party. (Which I support.)

Hard Reset’s stone-cold lock: Blexis at last. Talarico eclipses 51%, Paxton has one of the worst underperformances of this entire election cycle. President Trump lashes out at Paxton for a week straight. A handful of especially annoying pundits at CNN and Axios start asking if Talarico is a 2028 presidential candidate. He is not; if you’re the Democrats, you take this unfathomable W and keep a stranglehold on a Texas Senate seat for six years.

I read through it on Monday morning. The risk factors section of a filing like this tends to be the most interesting — Hamilton Nolan had a great piece this week walking through it — but my attention kept snagging on every mention of regulation and safety.

Anthropic has long made introspective pro-safety and regulation stances part of its brand identity. Iterations of the same statement appear all over their website, summed up by this one in Claude’s Constitution:

Anthropic occupies a peculiar position in the AI landscape: we believe that AI might be one of the most world-altering and potentially dangerous technologies in human history, yet we are developing this very technology ourselves. We don’t think this is a contradiction; rather, it’s a calculated bet on our part—if powerful AI is coming regardless, Anthropic believes it’s better to have safety-focused labs at the frontier than to cede that ground to developers less focused on safety. Anthropic also believes that safety is crucial to putting humanity in a strong position to realize the enormous benefits of AI.

This company has set itself proudly on a pedestal as a leader of safety as well as advancement. It publicly called for regulation and rallied around safety bills at the state level — preferable ones. Anthropic publicly opposed SB 1047 in California, an AI safety bill Governor Newsom vetoed in 2024, but they publicly endorsed the far weaker SB 53 that took effect this year.

It will be interesting to read any Frontier AI SEC disclosure, where they’re legally bound to tell the whole truth. The inevitability of this document that required thorough explication of AI’s safety risks was in the back of my mind while we watched Anthropic’s CEO affirm his own whistleblower last month. It was impossible to deny safety concerns with an S-1 draft on file confirming them.

It’s also good PR for Anthropic to be loudly self-aware of safety risks, and create tangible reference points (like supporting SB 53) to prove they’re serious about addressing them. It casts the company as trustworthy to a suspicious public. At the same time, it’s clear that Anthropic does not want AI regulation taken out of its own hands. Politico reported that representatives of Anthropic have been moving in silence on Capitol Hill to promote its own regulatory bill — it proposes government ability to block the release of a model if there was imminent risk of it killing “more than 50 people.” Which, wow. I think that says more than any way I could analyze it.

The company weighed the risk — included in the S-1 — of its political activity and advocating for regulation, and concluded it was a worthwhile one.

The political environment and our engagement with policymakers and public positions on AI regulation and safety could subject us to heightened scrutiny, political backlash, or disproportionate regulatory action.

Our commitment to responsible and safe AI development includes active engagement with policymakers, regulators, and other stakeholders regarding the development of AI-related laws, regulations, and standards. We have participated in, and will likely continue to participate in, political processes, including through contributions, advocacy, or support of political organizations. In addition, our executives and other representatives have publicly expressed and may in the future publicly express views on AI policy, regulation, safety, and related social or political issues.

Share

The risk, according to Anthropic, is opening itself to further scrutiny from the public or targeted retaliation from the government. President Trump has famously referred to Anthropic as an “out-of-control, Radical Left AI company,” and its relationship with this administration has been publicly tenuous — although CEO Dario Amodei was in attendance at Trump’s AI executives summit last month, and the two had a private dinner together beforehand where they reportedly discussed AI regulation. Specifics of that conversation were never made public.

A few other statements from the S-1 regarding safety, regulation, and Anthropic’s political activity:

• Concerns have been raised both within our organization and by other researchers, regulators, policymakers, and industry participants that sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity, particularly if such systems were misused, replicated, weaponized, or allowed to operate with a high degree of autonomy.

• As part of our political engagement, we have advocated for, and will likely continue to advocate for, regulatory approaches or industry standards that we believe promote the safe and responsible development of AI but that could increase our compliance burdens, limit our product development or deployment flexibility, or otherwise slow our growth relative to competitors.

• Any safety-related controversy or incident, whether actual or perceived, and regardless of fault or frequency, could prove significantly damaging to our business.

• Regulators have and could continue to use our disclosures as the basis for heightened scrutiny or more stringent requirements applicable to us compared to less transparent competitors.

• We have also advocated, and expect to continue to advocate, for regulation and oversight of frontier AI, including industry-wide standards primarily relevant to the use of AI to accelerate AI development, so that AI capabilities do not advance faster than the safety, alignment, and security practices needed to support them, and we have proposed working with the U.S. government and other frontier AI developers to establish safety standards. Regulatory or other measures of the kind we advocate, if adopted, would apply to us and could increase our compliance burdens, limit our product development or deployment flexibility, affect how and when we develop or release new models, or otherwise slow our growth.

• There is no assurance that policymakers will adopt the approaches we support, and our advocacy has in the past drawn, and may continue to draw, criticism, heightened scrutiny, or other challenges.

There are a few things that are especially interesting to me: the acknowledgement of existential danger was inevitable in this filing, and a reminder that Anthropic has more incentive than ethics to mitigate it; obviously, if an AI catastrophe reshapes the world, it reshapes the company, too. Anthropic acknowledges regulation as a potential risk to business and its advocacy as heightening that risk. And, Anthropic says even perceived instances of safety catastrophes could affect it negatively. Good PR matters, especially for an industry that’s largely mistrusted by the American public.

That all brings me to the coming election in which data centers and AI regulation have become mainstream issues. We at Hard Reset have been keeping an eye on Anthropic money in the 2026 elections; My colleague Alex Shultz has written throughout the month about recent political donations to conservative causes by Anthropic co-founder Tom Brown. I wrote about AI policy’s role in the San Francisco congressional race, in which millions in Anthropic corporate donations were funneled through a PAC to one of the candidates, Scott Wiener (who happens to have written the two aforementioned California bills). Because of the deniability political action committees provide, and the technicality of how money was moved around within the PAC, Wiener got away with his repeated defense of the donation: “There is no Anthropic money in this race.”

There is Anthropic money in races across the country, in large part via Anthropic’s corporate PAC of choice, Public First Action — which describes itself as “a bipartisan organization designed to educate Americans on key AI issues and advance an AI policy agenda supporting safeguards” — and the bipartisan candidates that PFA is using Anthropic’s donations to promote say they’re in favor of regulation.

A week from the election, Anthropic stayed consistent with characteristic pro-safety rhetoric in its S-1 filing. It allocated millions towards electing candidates of choice, very intentionally directing that money through political action committees. All the while quietly lobbying for its own regulation framework federally, with a track record of opposing strong safety measures and endorsing more moderate ones.

I’m not exactly comfortable with Anthropic’s altruistic motives for its political activity. Will Anthropic-financed candidates support or prioritize Anthropic-written bills? How much power over how AI is regulated in the United States will be determined by one frontier company saying it wants to mitigate existential risk, and what advantages and flexibility will it allow itself? Regulation is hardly regulation if a company wins its own preferences for external oversight.

And a public show of humanitarian proclivity is a nice bonus.

Share

A separate sobering study from last week reported that more than one in a quarter of Americans are facing medical debt, with one in six having more than $5,000 in medical debt. A particularly sad anecdote—though I suppose not entirely surprising for the American healthcare system—was from a young woman in Nebraska: “[One] patient, Maggie Williams, 28, a mother in Omaha, said she ended up with more than $20,000 in hospital bills she couldn’t pay after she required treatment last year for a high-risk pregnancy. She had qualified for Medicaid but lost eligibility, she said, after her husband’s income rose. While she was able to obtain coverage through his policy, she said, she still had the five-figure bill.”

With this all as the backdrop, the narrative is ripe for the claim that the efficiency of A.I. may save the doctor-patient relationship, the insurance industry—or even that one day, A.I. may even replace doctors and auto-diagnose or prescribe to patients who are desperate for quick and cheap care.

I spoke with Vamsi Varra, M.D., a dermatologist and machine learning scientist based in the Bay Area who has mixed feelings about the use of A.I. in medicine. He says that people who believe that A.I. will replace doctors “are deluding” themselves. Yet he reiterated that the issues in the healthcare system remain significant, including burnout of clinicians, and the top-down pressures from monied interests forcing the consolidation of practicing doctors—as well as an emphasis on patient volume rather than proper care. Dr. Varra is concerned that with A.I., the powers-that-be will use it as a tool to further consolidate power rather than as a remedy to solve many woes of the medical system.

This year, Dr. Varra started his own telehealth platform, in an attempt to ethically and asynchronously communicate with more patients without corporate medicine pressuring him to do so. Dr. Varra doesn’t use A.I. with his platform, mostly because patients have so little trust in its results.

“Patients don’t trust the quality of the A.I.—or what companies are going to do with the data they put into A.I. applications,” Dr. Varra told me. “I don’t trust it either.”

However, Dr. Varra has seen some interesting use cases for A.I. in dermatology, including a public research model that was trained, using predictive A.I., on over a hundred thousand images of benign and malignant skin lesions. Each image was meticulously labeled by a human dermatologist—allowing the model to learn and associate the pixels in the images with the association of being benign or malignant. While this model was confined to the research setting, Dr. Varra can see other potential A.I. use cases reaching real-world clinical deployment—including the automation of prescription acne topicals, because the risk of the wrong topical for retinoid acne is ostensibly non-existent.

Outside of dermatology, Dr. Varra posits that A.I. can be a helpful tool in getting certain test results for patients faster. For example, getting a blood cell count via blood smear can be very tedious when done manually; today this task is fully automated. He also points to the benefit of “ambient scribes” in taking notes and thereby preventing doctor burnout. However, this is not a panacea either: a recent study from the University of Edinburgh found “patients could be less likely to share personal information such as substance abuse history if they knew the conversation was being processed by AI.” Beyond that, there are many questions that arise as to where and how that data is being hosted and stored.

With the recent Elizabeth Holmes-Nathan Fielder film as a backdrop to our conversation, Dr. Varra is also concerned about the potential for A.I. hype in healthcare. To that end, Dr. Varra is actually in the midst of aiding the Utah Office of Artificial Intelligence Policy in reviewing an approved proposal for a pilot to automate acne prescriptions in Utah. It initially speaks of humans being involved in the validation process; however the documentation for the pilot—launched by Nolla Health and focused on assessing acne from patients’ photos—cites seemingly non-existent papers, which Dr. Varra thinks might be A.I. hallucinations.

Utah’s pilot program is not limited to dermatology, either: another pilot from Expect Fitness will offer “virtual pelvic floor therapy for women in the state to help with urinary incontinence, pelvic pain and postpartum pelvic floor problems.” Dr. Varra has since communicated his findings to the Utah Office of A.I. Policy. The office stated they communicated Dr. Varra’s findings to Nolla Health, who in turn sent the Utah office errata which will be updated on the office’s website.

Perhaps the one use case where some people—even skeptics—can see use for A.I. is in drug discovery, development, or novel treatment within healthcare. Some of that has been born out: last year, A.I. helped a doctor in Philadelphia repurpose drugs for conditions including rare and aggressive cancers, fatal inflammatory disorders and complex neurological conditions.

But drug development is of course different from A.I. being used in the actual administration of care via scheduling, for instance. A WIRED report describes the challenges nurses at HCA Healthcare face with Palantir’s automated scheduling system—including inexperienced nurses being scheduled at the same time, leading to the one experienced nurse being taken away from a critically ill patient to train the less experienced nurses. The same report speaks to mass burnout among nurses, and how they are unable to contest their schedules locally with their hospitals—instead having to reckon with HCA headquarters, the largest hospital chain in the United States.

One doctor reported to The Atlantic that “his hospital had enlisted A.I. to draft messages to patients on his behalf—sometimes producing output for his review that he described as ‘completely absurd.’”

And generally, the healthcare industry is prime for power consolidation, as it is increasingly becoming vertically integrated. Insurance companies like Aetna are owned by pharmacies like CVS—which also has its own pharmacy benefit manager (PBM), CVS Caremark, to negotiate opaque drug rebates with pharmaceutical manufacturers. Insurance companies like Cigna own specialty pharmacies, and gargantuan hospital systems own doctors offices. Many parties are often engaged in self-dealing circles—in cahoots with big pharma companies—that unnecessarily raise the prices of drugs for patients.

Astra Taylor, the activist and author of End Times Fascism, points out that A.I. may actually be costing the healthcare system more, because hospitals are deploying A.I. to overbill private insurers, and private insurers deploying A.I. to combat the hospital billing, leading to a war of the bots. Taylor says that this has cost Americans over $1 billion in excess healthcare spending.

So while some argue A.I. potentially could be used, in certain realms—to make things better for patients—in reality, it may actually be used as an excuse for cost-cutting and enriching just a few—namely the hospital, pharmacy, and insurance executives (and the makers of the A.I. products). To that point, the intrusion of A.I. into the healthcare system mirrors its enablement of power consolidation in literally every industry, including entertainment and media.

While Dr. Varra may disagree with someone like Astra Taylor on the unequivocal badness of A.I., he still believes the healthcare system needs a desperate overhaul. “When you look at the amount of independent physicians now versus a decade ago, you realize that we have an extreme consolidation of healthcare,” Dr. Varra told me. “Every single aspect of healthcare is very monopolized and consolidated; incentives don’t align with what patients or providers want. Beyond a few exceptions, strategies will be implemented for the sake of profit.”

Subscribe now

What else we’re reading…

Anthropic’s IPO prospectus describes a future “age of miracles,” replicating fantastical salvation language that others have used to describe A.I. and its benefits. (Of note and relation to this piece: Anthropic CEO Dario Amodei has claimed that Anthropic has the potential to “cure all diseases.”)

Former Yahoo CEO Marissa Mayer is getting heat for declaring that she had to ask her A.I., “Dazzle,” to help her think of three traits to describe her daughter.

This is a very cool chart from a Brown student mapping where data workers manually annotating A.I. are based.

Microsoft is offering a voluntary buyout to 60,000 employees—a whopping 27% of their workforce. It mirrors predictions that knowledge workers may see more voluntary buyouts from corporations as white-collar labor is devalued.

I traced Brown’s sparse political donations over the years—which were not to Republican candidates—and his very recent pivot, which coincides with Anthropic’s clammy efforts to get into the good graces of the Trump Administration. Chief among Brown’s donations: $250,000 to the Senate Leadership Fund, a Republican super PAC. His wife Michelle Valentine, a Silicon Valley investor and entrepreneur, also began donating to Republican causes this year.

The FEC only lists donations to federal candidates and PACs, meaning it does not include donations to gubernatorial candidates or localized races. I checked a bunch of statewide databases to see whether Brown is quietly cutting even more checks to conservatives. As it turns out, he is!

Brown recently donated $5,300 to the gubernatorial campaign of Marsha Blackburn, currently a U.S. Senator who’s running for statewide office in Tennessee. Despite her xenophobic and racist rhetoric, Blackburn attracts the support of Big Tech donors like moths to a flame. In the Senate, Blackburn serves on multiple tech subcommittees. As a candidate for governor, she’s indicated that Tennessee will continue to welcome more and more data centers. “Let me be direct: Data centers are going to be a good and important part of Tennessee’s economic growth, but we’ve got to be thoughtful about their placement,” she said in June.

Blackburn has scored donations from other Anthropic executives and employees, including ostensibly liberal ones. Brown’s $5,300 stands out for its timing: other Anthropic donations arrived during the same week in late June, implying a coordinated effort and/or a Blackburn fundraiser. At the time, Blackburn actually did need more money; she was locked in a tight primary against other conservative opponents.

But Brown’s donation was received on September 7, a month after Blackburn had already clinched the GOP nomination. Even a blue tsunami will not imperil Blackburn’s chances of becoming Tennessee’s next governor; Brown’s donation was for the love of the game (and a mutual love of data centers).

Interestingly, Brown appears to have dropped a larger donation—$25,000—to a group called Project Red TX, which says that it “has successfully recruited a robust network of candidates and seen dozens of county-level offices flip Republican!” The Washington Sun covered the group’s aims during the 2024 election: “Project Red TX has recruited hundreds of candidates for small local races in South and West Texas.”

I say “appears” because Brown’s listed occupation on the donation form is “engineer,” rather than what he usually marks down (“co-founder,” “chief compute officer,” and “executive”). That said, I couldn’t find another Thomas B. Brown—his correct middle initial included—who works at Anthropic and has a San Francisco zip code. I certainly don’t think $25,000 is a typical donation for a regular ol’ engineer, even a well-compensated one. Brown is also literally an engineer. He’s just advanced a few notches above that in his career, and should probably list a more senior title on his donation forms if he’s going to wade into local Texas politics.

Is a Tech Billionaire Capable of Buyer’s Remorse After Ramaswamy Ads?

Ratmir Timashev isn’t world-renowned, but he’s a very wealthy man. He grew up in the USSR, then immigrated to the United States; in 1996, he obtained a master’s degree in chemical physics at Ohio State University. He cofounded a few software and data management companies, one of which sold for $5 billion. He’s now in the venture capital and AI business, and has a reported net worth of $5.4 billion.

Timashev strikes me as the sort of tech billionaire who prefers to keep a low profile. He’ll do magazine interviews to talk about his philanthropic interests, or perhaps to advance his business goals, but he’s not on X or Instagram. His LinkedIn posts consist of purposely bland hiring announcements and boring leadership advice. Last week, he wrote a guest column for The Columbus Dispatch about a research facility at Ohio State that will be built in his name. (He’s given more than $110 million to his alma mater.)

I briefly mentioned Timashev’s backstory in July, during another political donations roundup for Hard Reset. At the time, he’d just given $5 million to a super PAC called V-PAC. The super PAC supports one candidate: Ohio gubernatorial candidate Vivek Ramaswamy.

You might have read about Ramaswamy’s “October surprise,” by which I mean his campaign ran one of the most repulsive attack ads in the history of modern American politics. The ad features the mother of Ramaswamy’s opponent, Amy Acton, even though Acton and her mother haven’t spoken in decades. The reason for the estrangement, according to Acton, is that her stepfather repeatedly raped her when she was a child, and after he was arrested, Acton’s mother took his side, not her daughter’s side. In the attack ad, Acton’s mother repeatedly refers to her own daughter as a liar.

The Ramaswamy campaign’s closing message has managed to alienate a broad coalition of conservatives. Ohio Governor Mike DeWine, Ohio Senator Jon Husted, Iowa Senate candidate Ashley Hinson, and former presidential candidate Nikki Haley (who hates Ramaswamy’s guts) all publicly disavowed the ad.

Which brings me back to Timashev, who gave another $5 million to the Ramaswamy super PAC in mid-September. Other than Jeff Yass, who’s unrepentant about his pet causes and massive political donations, no one has given more to V-PAC than Timashev. V-PAC did not make the rancid ad mentioned above—I want to be very clear about that!—but it’s produced other gross ads, including one that alluded to Acton’s outpatient therapy services, which she says she received because she was sexually abused as a child. And again: the super PAC has no other purpose besides supporting Ramaswamy. It’s not divvying up donations to a bunch of other candidates.

FEC records and Ohio’s statewide database do not show other political donations from Timashev during this election cycle or prior election cycles, for that matter. It’s Ramaswamy’s super PAC and that’s it. Timashev’s track record is pretty apolitical. (At least, as much as that’s possible for a tech billionaire.)

Mindful of that context, I thought there was at least a small possibility Timashev would express regret about wading into the political realm for Ramaswamy. I reached out to spokespeople for two of Timashev’s companies—OH.io and EverWorker—and also sent a message to his foundation. I asked why he’s given a total of $10 million to V-PAC, if he thinks the Ramaswamy campaign’s ad is appropriate or not, and if he stands by his support for Ramaswamy.

To be fair to the spokespeople, I’ll note that I emailed on Saturday afternoon, which is not an ideal time to make a request for comment. But also: I gave everyone more than a day to reply, and if they’d asked for an extension, I would’ve granted it.

Unsurprisingly, I didn’t hear a peep from anyone. I will update this post if a response ever arrives. It seems to me, though, that Ramaswamy is going to lose on Nov. 3, and his desperate efforts to impugn Acton will only cause collateral damage to his biggest backers—Timashev included.

California Wealth Tax Antics (and a Sad Prediction)

Building a Better California, the Big Tech PAC that’s hellbent on killing California’s proposed wealth tax, is raising money at a genuinely unbelievable rate. Just this weekend, the PAC recorded $28 million in contributions. I am not the first, nor will I be the last, to point out that this endless spigot of funds could simply be subject to taxation.

There have been some signals that billionaires are preparing for the worst (paying a one-time 5% tax). Last week, the New York Times reported that Mark Zuckerberg threatened to move out of California unless Meta’s board of directors figured out how to “pre-emptively restructure his stock holdings.” Zuckerberg was apparently worried about his voting shares being diluted as a result of him selling stocks to pay for the wealth tax. Don’t worry folks: Meta acquiesced to his power play. Zuck’s iron grip on Meta remains.

My best guess, which is based on limited polling and me putting my pundit hat on, is that the wealth tax ultimately will not be enacted. If Prop. 40 (the wealth tax measure) passes, it can still be undone by Prop. 41 and Prop. 42, which are basically poison-pill measures that billionaires snuck onto the ballot. In the event that either proposition accumulates more votes than Prop. 40, the wealth tax is invalidated. That gives California billionaires three different outs. I do not like those odds, especially coupled with a lack of enthusiasm for candidates at the top of the ticket. (Xavier Becerra is not going to be among the Democrats who outperform expectations next week.)

To give you a sense of how much this ultimately meager wealth tax has turned into an existential threat among oligarchs: Frank Slootman, the billionaire chairman of the board at Snowflake, recently gave $5 million to Building a Better California. A nice chunk of change, certainly, but that’s not the interesting part. On the donation form, Slootman’s address is in Florida, not California. And he’s registered to vote in Montana, according to the state’s voter rolls. This certainly does not preclude him from owning a property (or a bunch of properties) in California, but it’s a fun reminder that some of the billionaires who are most frightened of Prop. 40 aren’t even California voters—or living in the state full-time.

A few more October donations:

• Accounting for another two weeks of donations during the first half of the month, Elon Musk has reportedly contributed roughly $420 million during these midterms. That’s more than he contributed during the 2024 election cycle, when he completed his transformation into a far-right megadonor.

• On Oct. 14, WhatsApp co-founder Jan Koum contributed $265,800 to a super PAC that supports Mike Rogers, Republican candidate for Senate in Michigan.

• Though it pales in comparison to some of his seven-figure donations, it’s worth noting that angel investor and Ripple co-founder Chris Larsen tossed another $250,000 at a super PAC for Senate Democratic candidates.

• LinkedIn founder Reid Hoffman gave another $5 million to a super PAC that supports Texas Senate candidate James Talarico.

• Airbnb cofounder and U.S. Chief Design Officer Joe Gebbia gave $1 million to a Ken Paxton super PAC.

• The Winklevoss twins continue to match each other’s donations, which is adorable. They each gave $125,000 to a super PAC that boosts House Republican candidates.

She didn’t know I was in the middle of annotating the startling report published last week by a coalition of research and labor organizations, titled Bossware Comes To California.

Her boss was innocuously and visibly looking on as she clicked between spreadsheets; what the bossware report describes is the discreet systemic implementation of AI-powered surveillance across sectors of labor. It names dozens of platforms designed to collect and digest data about individual workers, often discriminatorily, and often without employees’ knowledge or consent.

Workplace surveillance and automation has been an issue for decades, but the implementation of AI bossware systems has accelerated employers’ ability to intensify work, increase control over workers, and extract more value from their labor. Bossware, according to the report, can look like physical devices — GPS trackers, cameras, microphones, wearable or handheld devices — or electronic monitoring software collecting data from a worker’s performance to their “biometric information, such as fingerprints, facial features, eye movement, gait, or voice characteristics.” Some bossware systems continue to track and gather data on employees outside of work hours.

California recently passed four first-of-their-kind laws establishing beginner guardrails for automation in labor. SB 947 banned employers from using AI to unilaterally discipline or fire employees with no human oversight or corroboration. SB 951 required expanding research and the information available to workers when AI causes layoffs. AB 1331 banned AI surveillance in workplace bathrooms.

And AB 1883, the “Stop AI Strip Searches Act,” banned bosses from using AI surveillance tools to track, recognize, or predict the emotional states of workers, or collecting their neural data.

All of those measures should absolutely be in place, although the report emphatically states they are not comprehensive enough. The details, however, should make you pause and consider why they were necessary.

The Bossware report lists evidence compiled from the firsthand experience of twenty-five California workers across many industries, including healthcare, warehousing, software engineering, janitorial services, sanitation, and hospitality. The takeaway: Bossware is widespread across industries, especially surveillance that collects biometric data.

The Los Angeles Alliance for a New Economy (LAANE) is a part of the coalition that published the report. LAANE’s Executive Director, Víctor Sánchez, agreed that, of the report’s revelations, the secretive tracking of workers’ emotional states stands out.

“I feel like, honestly, it’s one of the most scary things that we kind of learned in doing this report,” he told me. “I think my mind goes immediately to,, what allows a technology to be able to make a determination on someone’s emotions. And what that takes is facial recognition, eye scanning, these things that are terribly intrusive. And so I think, for one, it’s scary to think what that tech could be used for in the future.”

The future of work is the thematic concern; the tangible ones — how the well-established encoded bias of AI is violating workers’ civil rights, how the use of bossware erodes worker power and enables union busting, how algorithmic management pushes workers to perform at unrealistic or impossible standards that endangers their mental and physical health — are glimpses of an anti-worker future. Automating human discernment and complex work will eventually deskill the human workforce and reshape labor entirely.

“This idea that workers themselves are being pulled out of decision making roles where they can exercise judgment — where they’re not analyzing, diagnosing or trying to assess critical problems — Instead, those processes are being automated,” Sánchez told me. “Largely an excuse to be able to pay workers less.”

He said that the conversation of how AI and automation could change the workforce is no longer a conversation; it’s happening, and the data employers are collecting from their workers is helping accelerate it. “The harms are being realized now,” Sánchez emphasized. “ I think the purpose of these tools is to deskill, to move towards this workerless future, regardless of whether or not the tech is actually good enough to do what they want it to do. I think for all intents and purposes, employers are deploying this technology because they see it as an important stepping stone to get to where they want to go.”

Share

The discrimination enabled by uncritical use of biased AI systems in the workforce is another alarm sounded in the report. Especially where biometric information and facial recognition technology is concerned, racial bias in AI detection systems is well-established. Bossware isn’t even close to an exception. Workers of color and immigrant workers are disproportionately targeted for discriminatory and predatory employer practices bossware systems enable. ‘Surveillance wages,’ or employers’ use of predictive analytics to identify the lowest pay a worker would be willing to accept, is only one example.

“These tools are imperfect, and so oftentimes when they’re treated as gospel, the outcomes are thought to be perfect when they’re not,” Sánchez expanded. “We’ve got some examples in the report that we uplift, of workers being terminated, [and] having no course of action because it’s not transparent who it is that they can speak to. And so these become tools of furthering inequities in the workplace.”

When it comes to a course of action and call for regulation, the report says plainly that they don’t expect federal regulation of bossware in the near future, but discourages nihilism. The four California laws aren’t a fix, but they are a win, and an established regulatory baseline. The report lays out four new policy points as immediate goals.

1. Employers, not algorithms, should be the ones making decisions about workers, and workers must have the right to appeal.

2. Workers should have the right to know when and how bossware is used by their employer, and the right to access and correct their data.

3. Employers should be prohibited from using bossware in ways that are intrusive and harmful.

4. All workers must be protected, and they should have the power to enforce their rights.

LAANE and the coalition behind the report, Sánchez told me, is working on local regulatory frameworks in four major regions of the state, with the hope of any eventual federal regulation having California as an example to build upon.

“And a lot of those have to be, I mean at its core it’s about democratizing the process, right? It’s making sure that people are at the center of how these technological tools are being deployed and not left out on the periphery,” he said. “We think that that can create great momentum, not just at the state level, but across the country, just given the sheer size and scale of some of our regions. LA alone is larger than a lot of states.”

Starting at the local level for something as ubiquitous as workplace AI surveillance regulation seems frustrating, at first. Ideally, this would be shut down at the federal level very soon, and NLRB standards and worker protections could evolve as quickly as AI. Imminent federal progress being unrealistic doesn’t make local efforts futile. Sánchez reiterated that.

“I think city council members, even school districts, right, have an opportunity to look at how AI and tech, or tech that’s AI enabled, is being deployed within their jurisdiction, and look for how they can meaningfully put up practical guardrails to make sure that people are at the center,” he said. “I think, you know, a large part of the challenge with state legislation is, there are a lot of interests that are at play. And so we weren’t able to win everything that we wanted to.

I do think that it’s got to be a joint effort between local and state, because local is really where the rubber meets the road. I mean, you’re seeing a lot of the fights around data centers emerge locally. A lot of the stories that we’re hearing are stories that people are experiencing on the day to day, and that has to ultimately guide the type of change that we’re trying to win, how people are being impacted every day.”

No matter what you do for work, and even if you are unaware of it happening, it is very possible you are being tracked, data mined, and quantified by bossware powered by imperfect AI systems that may or may not be overseen by humans. As far as what’s possible with our data employers collect — is it worth finding out?

“We’re looking at the deployment of bossware as a kind of laboratory experiment,” Sánchez told me. “Now there’s obviously a clear, longer-term threat in terms of where these companies want to go.”

A few months later, this CEO’s tech startup—where I was employed—ran out of cash. He had lied about how much investment he had received, claiming he had hundreds of millions of dollars more than he did. And he was spending recklessly, all while telling everyone that a bridge loan was coming in to keep things afloat.

Two years later, after the dust settled, I confronted him about the actual amount that was raised. He said that the journalist who had reported the incorrect fundraising number had simply misheard him, and that he’d forgotten to correct it. He said it with such confidence that I almost wanted to believe him.

All of this came back to me after I watched Nathan Fielder’s incredible film You Can See Everything, which examines the question of who Elizabeth Holmes—the ex-CEO of blood-testing startup Theranos, which defrauded investors and patients—actually is.

The premise of the film involves experimental comic and performance artist Nathan Fielder staying at Elizabeth Holmes’ rented beach house before Holmes serves an eleven-year prison sentence for fraud. During the three hour film—which felt closer to maybe an hour—viewers are whisked through an emotional rollercoaster, one that frequently challenges the perspectives you believe to be true, even within the duration of the film.

At the beginning, we see Holmes is a new mother, almost in trad wife form. She is breastfeeding, going on beach walks and swims with her fiancé Billy Evans, and attempting to whisk eggs. At several points, preparing for her impending prison stay away from her children and partner, she is in tears. You feel a bit bad for her.

In one scene, Holmes—called “Liz” by both her fiancé Evans, and Fielder—is shown the famous Abbott and Costello skit “Who’s On First?”

Fielder shows her the scene after Holmes insists that she didn’t understand why the words “this product can work,” in relation to her failed blood-testing kit, could have been misleading as to whether her product actually works. Fieler is showing her “Who’s On First?” as an absurd demonstration of how the sequence of words and the intonation of language matter in how people interpret them.

Liz watches and laughs a bit at the skit. But when asked by Fielder whether she understood why it was funny—and what the baseball players’ names are—she said she didn’t know (that “Who’s” on first, “What’s” on second, and so on). Fielder is skeptical that Liz is not aware, as are we. But we remain astounded and bemused, I think, by her doe-eyed commitment to her own bit.

At another point during Fielder’s stay, he has Liz watch actress Amanda Seyfried’s performance of Holmes in The Dropout—the fictionalized television show about Theranos. Liz watches, but then snuggles up to her fiancé in tears. Eventually she implores Fielder to turn it off.

Now, while Fielder and the viewers remain skeptical of Liz’s feelings—especially given Liz’s declaration to Fielder, at one point, that she has never lied—the first part of the film elicits a bit of sympathy for a weeping Liz. And while she’s in prison, our sympathy intensifies.

We begin to believe that the true villain in the story is Liz’s fiancé, Evans. In the second part of the film, Evans becomes almost belligerent compared to his prior self. We learn that he has power of attorney over Liz, and that he is a Charlie Kirk fan. We discover that he is posting on X.com as “Elizabeth Holmes,” unbeknownst to her—as well as negotiating monetary deals on Liz’s behalf related to the Fielder’s documentary. All of this matches the narrative that Holmes, during her Theranos years, was a victim of Sonny Balwani—the much-older man who serves as Holmes’ COO at Theranos.

In the second part of the film—because Fielder cannot access Holmes by video in prison—he actually enlists actress Amanda Seyfried to play Holmes again, in various scenes that are scripted based on real life. Seyfried the person eventually becomes the voice of reason in the film. She challenges Fielder when he claims that Liz is truly in love after seeing her embrace Evans in the prison visiting room, reminding him that he may be biased in his endearment. The film then cuts to a scene of Nathan Fielder frolicking in the ocean with Liz before she goes to prison (reminding us that he could be a biased narrator, or that he is intentionally becoming a biased narrator to remind us of that potential.)

Psychologically, the film is brilliant. It breaks our boundaries of what we think is reality—over and over and over. If Holmes is manipulating Fielder, is she knowingly doing so? Or is she so coded and predisposed to manipulate as a form of survival that it genuinely feels authentic to her? Is she crying real tears? Does she actually miss her children—or is it a bit to get a pardon?

The documentary, perhaps for artistic reasons, doesn’t get into the history of Holmes’ fraud at Theranos. That was intentional, I think—because it parachutes us into a world where we can see ourselves being defrauded by a context-less Holmes, even if only for a moment.

It also appeals to the very human sensibility we have to want to see the best in people—especially manipulative people who know how to push us to see the best in them. And because fiancé Evans serves as a villainous foil to Liz’s seemingly pure motherhood, we have a new “villain” —making it very easy to look past the fact that Holmes may even be manipulating Evans, knowing that she will let him believe he is manipulating her.

In an abusive or manipulative relationship, one may only be able to see things with clarity after separation. And I must admit that it actually took me a day or two to remember that Holmes does in fact know what the word “can” means in the context of a tech product’s potential. It took me a few conversations with others who have seen the film to remember how absurd it is to think Holmes has no idea what is going on around her.

As Theranos whistleblower Tyler Shultz reminds us in an op-ed for STAT News: “The device [Holmes] showed Fielder on the kitchen table was not the reason she went to prison. Elizabeth showed him a future version of the Theranos product that was never completed and never used in order to lead Fielder down this BS semantic rabbit hole and to trick him into litigating the entirely wrong thing.”

Right after my encounter with the fraudulent tech CEO in 2016, Donald Trump was elected. At that time, it made me wonder—with fear and some sadness—whether it was indeed true that manipulation is just something that weaves the fabric of reality.

I think a part of Nathan Fielder’s takeaway is that we are, sadly, used to expert manipulation—especially in a post-truth world, and particularly when it comes to the people in extreme power. Because don’t we want to believe that our leaders are benevolently creating investment accounts to benefit children—or graciously giving shares of their trillion-dollar company to children? Isn’t it human to believe it’d be nice to rely on someone’s “genius” or altruism? Isn’t it natural to want to feel like someone is genuinely sad about going to prison, or that an endeavor to scan for disease via blood test was all a misunderstanding?

The whole film may be manipulative for everyone involved. But it does give us a glimpse into the psycho-pathology of Holmes, and those like her. While you can’t see everything yet in this film about the minds of the tech oligarch elite, it’s a start.

Last week, New York City Comptroller Mark Levine sent Anthropic a letter asking it to “pause” its initial public offering (IPO), which is currently expected next month. Levine’s letter points to Anthropic’s “unusually direct public statements acknowledging that the pace of frontier-model capability development may now exceed society’s capacity to understand and control the resulting systems” as the reason for delaying the company’s stock market debut.

I’m not here to litigate whether those Anthropic statements should be taken seriously or dismissed as a stealth marketing campaign designed to showcase the supposed power of its AI models (though, more on that later). Instead, I want to focus on a separate issue: there is strong reason to distrust the information the company provides the public regarding its finances, and that provides a clear legal basis for the Securities and Exchange Commission (SEC) to block the company’s planned IPO. Allowing Anthropic’s massive IPO to go forward would threaten to further destabilize capital markets that are already dangerously dependent on the speculative and increasingly teetering AI trade.

Share

Publicly traded companies must release audited financial statements every quarter showing their actual revenues, expenses, and other key financial information. Privately held companies, even if they have trillion-dollar valuations and hundreds of billions of dollars in off-balance-sheet obligations, need not do so. Anthropic, like many Silicon Valley startups, has taken advantage of that loophole by occasionally and selectively leaking information about its finances to reporters, most notably a metric called revenue run rate.

Run rate is a dubious projection that takes a company’s revenues over a recent period and calculates what the company’s revenues would be if it sustained that performance for a full year; it does not reflect the company’s actual annual receipts. That enables a company to, say, wait until it has an unusually lucrative month, multiply that month’s revenues by 12, and represent the resulting figure as its “annualized” revenues — even if the company has not made close to that amount of money in any actual 12-month period. For example, Anthropic announced a run rate of $9 billion at the end of 2025, but when forced to disclose its actual revenues in a court filing, the company revealed that its revenues were closer to $5 billion over the course of 2023, 2024, and 2025 combined.

Because run rate is non-standardized, it is difficult for outsiders to know what an announced run rate includes. That can be a boon to companies with shaky finances because it allows them to project an artificially rosy financial picture by cherry-picking when and how it calculates and announces its run rate. But the past couple weeks have shown how this gambit does more to mislead than inform as well as how run rate announcements can backfire when a company’s revenue growth slows.

The first illustration of that came in late September, when some news outlets reported that the run rate of OpenAI, Anthropic’s archrival, was approaching $70 billion. But earlier this month, the Financial Times reported that the ChatGPT maker’s run rate was closer to $50 billion — nearly 30% lower. The discrepancy apparently arose, as Bloomberg reported, because “Anthropic counts as its revenue the gross amount of certain sales through cloud partners like Amazon, while OpenAI recognizes only its share of revenue generated from such partners” (emphasis mine), and someone failed to account for that distinction when trying to make an apples-to-apples comparison.

The italicized bit should have received more coverage than it did. At best, it shows that run rate basically means whatever a company wants it to mean and that model developers’ incestuous relationships with cloud providers make it easy for them to manipulate the financial information they report to the public. The incident starkly illustrated how little markets know about the actual financial performance of the two companies whose performance will ultimately determine whether the multi-trillion-dollar AI build-out generates enough revenue to cover its costs.

Less than a week later, another Financial Times report revealed that Anthropic’s revenues for the third quarter came in at $16.8 billion, or $67.2 billion on an annualized basis. Early news coverage spun the figures in a positive light, emphasizing that Anthropic’s revenues had grown 50% from the previous quarter as well as the fact that Anthropic told investors that it had achieved a profit “on an adjusted basis.” The best analysis of the latter claim came from the FT story’s comments section:

Shame on FT for printing that claim without pressing Anthropic to specify what the “adjusted” profitability calculation includes and excludes. But the bigger issue is that Anthropic told investors (again) that it is “profitable” when it almost certainly was not (today, the Wall Street Journal reported that Anthropic actually lost a staggering $81 billion in the first half of 2026), and that someone — most likely one of Anthropic’s investors or financiers, if not Anthropic itself — used the press to communicate the clearly misleading “profitability” claim to the public.

But despite these efforts at positive spin, the financial commentariat soon pointed out that the $16.8 billion revenue figure actually showed that Anthropic’s revenues had effectively stopped growing. That is because Anthropic had announced a revenue run rate of $65 billion as of the end of July, which comes out to $16.25 billion per quarter. Had its revenues continued to grow quickly over the next two months, one would expect its Q3 revenues (covering July, August, and September) to be substantially higher. Instead, its full Q3 revenues came in just 3% higher than its July run rate.

Perhaps there is an explanation for this that will allow Anthropic to shore up the deteriorating perception of its finances. But regulators should not be fooled. The most generous interpretation of the leaked information regarding its revenues and profitability is that Anthropic’s finances are so opaque that even sophisticated investors and financial journalists have trouble deciphering them. The more ominous possibility is that Anthropic and/or its financiers are actively attempting to mislead the public regarding the model developer’s financial condition.

Subscribe now

That provides a more-than-sufficient basis to block Anthropic’s IPO until the public can get a clear and reliable picture of its financial condition.

Before a company can go public, the Securities and Exchange Commission (SEC) must approve its registration statement. The SEC can delay or block an IPO if it believes a company’s registration statement, which includes information on the company’s finances, is “on its face incomplete or inaccurate in any material respect.” SEC guidance states that the regulator “will not declare a registration statement effective if the staff has reason to believe that the disclosure is incomplete or inaccurate in any material respect.”

There is, at this point, little reason to trust and every reason to doubt any financial information that Anthropic provides to the public. The SEC should refuse to approve its registration statement until it provides two consecutive quarters’ worth of audited financial statements. Ideally, given the possibility that Anthropic or those acting on its behalf are attempting to mislead the public regarding its finances, the SEC should demand that those disclosures do not merely match but exceed the financial transparency expected of public companies.

Anthropic’s recent musings about the potential for its technology to destroy civilization provide an additional reason for regulators to delay its IPO, regardless of the motivations for those statements. If senior researchers and executives at Anthropic sincerely believe that there is a significant risk that the technology upon which its entire business rests could escape their control and cause human extinction, then the company is too dangerous to exist, much less be given access to public capital markets. If, on the other hand, the statements are a disingenuous publicity stunt, that deception provides still further reason to distrust the information in the company’s registration statements.

The economic consequences of letting Anthropic’s IPO proceed now could be serious, particularly with OpenAI gearing up to go public in the coming months as well. Stock markets are already historically overvalued relative to their earnings and disproportionately concentrated in a handful of tightly interconnected mega-cap companies from the tech sector. That has already created a risk of a historically damaging stock market crash; never before has such a huge amount of wealth depended on the performance of a small number of companies whose fortunes are inextricably linked. Adding another huge, unprofitable tech giant (much less two) to public exchanges would exacerbate both of those factors and create a higher risk of contagion if (or, rather, when) Anthropic reaches a financial precipice.

The SEC has the authority to stop that from happening. The agency should use it.

Since July, Musk has given at least $276 million to boost Republicans’ midterms odds. A significant chunk of those donations have been funneled to a super political action committee (PAC) that’s supportive of Paxton. One could argue Musk is attempting to rig the Texas Senate race. Not me, though. I think it’s awesome that with the snap of a finger, an on-again, off-again trillionaire can ensure that a Paxton super PAC raises almost triple the quarterly funds of James Talarico’s campaign committee and main super PAC combined.

Musk’s oligarchical output is the biggest takeaway from this most recent Federal Election Committee (FEC) quarterly filing deadline. Proportionately, he is unmatched. Sometimes, though, the same super PAC filings reveal donations that fill in important blanks and tell their own compelling story.

Anthropic CEO Dario Amodei has created headaches for his AI company by butting heads with the Trump Administration and the Pentagon. (My take, as I’ve written about at length, is that Amodei is correct to invoke use-of-force red lines for his AI tools, but he does not go nearly far enough, and in fact has agitated for an even more hawkish stance against China, the world’s other superpower.)

To compensate for Amodei’s annoyingly squishy liberal vibes, Anthropic has resorted to back-channeling with the White House via Tom Brown, another company co-founder. Brown has reportedly quelled Anthropic’s testy relationship with both the White House and Musk. When Trump recently hosted a bunch of tech oligarchs so they could agree on one big beautiful AI safety pinky promise, Brown was in attendance. Amodei, too! Trump even said he likes Amodei now.

What’s Brown’s secret sauce? He’s tall—six-foot-six—and Trump is on the record liking tall and beefy men. But according to a new Wall Street Journal profile, the real key to Brown’s schmoozing is that he’s a Republican. “Brown bought a house in Washington earlier this year with his wife, a startup co-founder and former venture investor named Michelle Valentine,” the Wall Street Journal reported. “The couple has said in private conversations that they are allies of the Republican party and want to establish relationships in Washington, people familiar with the matter said.”

The phrase “allies of the Republican party” got my attention. It’s an endorsement, but a lukewarm one. (Another sentence in the profile refers to Brown as a Republican, though it’s not attributed to anyone.) I figured now—just after the third-quarter FEC filings deadline—was the perfect time to look into Brown’s history of political donations. What I found is less GOP stalwart, more tech billionaire who’s making a conveniently timed rebrand.

A brief PACs primer before I go any further: A campaign’s principal committee can accept a “maximum” donation from an individual of $3,500 for a primary, then $3,500 for a general election. Super PACs can raise unlimited funds, but are not supposed to coordinate directly with any candidates—a farcical and easily exploitable system. There are joint-fundraising committees that have higher donation limits than a campaign committee. There are also dark-money groups that don’t report their donors, just their aggregate donation amounts, because they are registered as nonprofits. The FEC barely investigates anything anymore because it has four vacant seats out of six possible commissioners.

In 2016, Brown listed his profession as “not employed.” He was nowhere near his current estimated net worth of $15.5 billion. Guess who received $35 of his hard-earned money? That would be Sen. Bernie Sanders. So…probably not a Republican a decade ago. Fair enough. Many such cases among the tech elite. According to FEC filings, Brown didn’t log another donation to a federal candidate until 2021. This time, he donated $250 to a longshot congressional candidate in Virginia named Victoria Virasingh. She previously worked at Palantir, so I’m not going to declare her part of the Sanders faction of the left. She did, however, run as a Democrat who said she wanted to model her politics after Squad member Rep. Ayanna Pressley.

In 2022, Brown gave $2,900 (the maximum allowed during that election cycle) to Carrick Flynn, a congressional candidate in Oregon who garnered the support of ultra-wealthy figures in the effective altruist community. Again: not exactly a leftist, and Flynn certainly didn’t have a left-leaning base of support. But he did run as a Democrat! For those keeping track at home: three political donations for Brown, none to Republicans. (To confirm Brown’s identity, I cross-referenced the address he listed in 2016 with the address he listed on later forms, when his employer was Anthropic.)

In April of this year, Brown’s wife Michelle Valentine—a Bay Area entrepreneur—started donating to Republicans. I couldn’t find any previous donations that she sent to federal candidates. If you’ll recall, April was smack-dab in the middle of Anthropic’s Cold War with the Trump Administration. In late February, the Pentagon announced it was designating Anthropic as a supply-chain risk. In March, Anthropic sued the Pentagon over that designation. In mid-April, Amodei met with senior White House officials. “The company appears to be inching back into the White House’s good graces,” CNBC reported afterwards.

That same month, Valentine logged more than a half-dozen “maximum” donations to House and Senate Republicans. In May, Valentine gave $5,000 to the hilariously named Making a Responsible Stand for Households in America (MARSHA) PAC; it dispenses funds to allies of the far-right Sen. Marsha Blackburn, who’s running for governor of Tennessee. I also spotted donations to Louisiana Sen. John Kennedy and Alabama Sen. Katie Britt, among others. More recently, Valentine cut a check for $157,900 to a joint fundraising committee for Senate Republicans.

And wouldn’t you know it? Brown is now entering the fray as a Republican donor. In July and August, Brown maxed out donations to Rep. Mike Rogers (who’s running for Senate in Michigan against Abdul El-Sayed), followed by Rep. Anna Paulina Luna (who believes the 2020 election was stolen and that there should be a Charlie Kirk statue in the U.S. Capitol). On the last day of the most recent quarterly filing period, the Senate Leadership Fund—a Republican super PAC—marked down a $250,000 donation from Brown. He also appears to have given $75,000 to the joint fundraising committee of Mississippi Sen. Roger Wicker, who’s the chair of the Senate Armed Services Committee.

Taken in totality: After years of rarely giving to federal candidates, Brown and Valentine have donated hundreds of thousands of dollars to Republicans and conservative causes during the latter half of 2026. I expect they’ll eclipse seven figures in donations by Nov. 3.

It’s possible Brown was always right-wing, despite his prior, sporadic donations to Democrats. I doubt it, though. It’s more plausible that Brown was radicalized over the last few years, same as countless other tech billionaires who want to take advantage of MAGA’s distaste for regulation and whatever is considered woke at the moment.

I’m not discounting a third possibility: that Brown has drifted right, sure, but he and his wife also see a golden opportunity to suck up to MAGA. After all, Brown’s billionaire net worth is dependent on Anthropic’s (likely) impending IPO. It’s in his best interest for the Trump Administration and a Republican-led Congress to be Anthropic’s allies, not their enemies. One surefire way to do so is to drop some well-timed donations that are meaningful enough to get on Republicans’ good side, but not huge enough to garner wall-to-wall coverage a la Elon Musk.

Here are other donations from the tech industry that caught my attention:

• Marc Andreessen maxed out donations to a bunch of Republican Senate candidates, including Rep. Julia Letlow, who’s running for office in Louisiana to replace outgoing Sen. Bill Cassidy; Sen. Darline Graham Nordone, who’s trying to hold onto her deceased brother Lindsey’s seat in South Carolina; Rep. Ashley Hinson, who’s running to replace retiring Sen. Joni Ernst in Iowa; Alaska’s incumbent Sen. Dan Sullivan, who’s in a tight race against Mary Peltola; West Virginia Sen. Shelley Moore Capito; and Ohio Sen. Jon Husted, who’s a narrow underdog against Democratic opponent (and former Senator) Sherrod Brown.

• Andreessen’s business partner Ben Horowitz also donated to South Carolina Sen. Graham Nordone, Alaska Sen. Sullivan, West Virginia Sen. Capito, Ohio Sen. Husted, Iowa Rep. Hinson. And he sprinkled some cash to the DCCC ($22,150) as well as to Rep. Seth Moulton, an unsuccessful Senate candidate in Massachusetts.

• Anduril CEO Palmer Luckey gave $36,900 to a GOP joint fundraising committee, an amount exceeded by his wife (she gave $61,800 this quarter).

• Reid Hoffman, the billionaire LinkedIn founder and acquaintance of Jeffrey Epstein, gave $5 million to the Senate Majority PAC, a super PAC for Democratic Senate candidates. He tacked on $2.5 million for a handful of other super PACs backing left-of-center causes and candidates.

• Facebook co-founder Dustin Moskovitz made his biggest public donations since the 2024 election: $4 million to the Senate Majority super PAC, and $4 million to the Future Forward super PAC (another bottomless well for Democratic Party megadonors). Moskovitz’s wife, Cari Tuna, matched his donations.

• Netflix co-founder Reed Hastings dispersed $1 million to the Democratic Senate Majority super PAC, $1 million for the James Talarico’s main super PAC, and $300,000 to a Hakeem Jeffries joint fundraising committee.

• Palantir CEO Alex Karp donated a combined $515,000 to two super PACs that support conservative war hawks.

• Angel investor Chris Larsen contributed $2 million to the Democratic Senate Majority super PAC.

• Airbnb co-founder Joe Gebbia, who also holds the made-up sounding (but real) title of U.S. Chief Design Officer, ponied up $1 million for a MAHA super PAC.

• Steve Ballmer is still in hiding after his Los Angeles Clippers were severely punished by the NBA for salary cap circumvention violations. Connie Ballmer (Steve’s wife) gave $3 million to a gun control PAC.

• The Winklevoss twins combined to contribute millions of dollars to the MAGA Inc. super PAC. I am having trouble tallying the exact figure because, no joke, it appears they made the contributions via Bitcoin? Didn’t know you could do that. Very on-brand for those fellas.

The U.S. House seat representing San Francisco, the global capital of AI, is without an incumbent for the first time in four decades; Speaker Emerita Nancy Pelosi is retiring after 20 terms in Congress. It’s a very tight race to replace her between California State Senator Scott Wiener and San Francisco Supervisor Connie Chan. Wiener, with endorsements including the San Francisco Chronicle and California Democratic Party, finished first in June’s primary. Chan came in second, weeks after Pelosi endorsed her. Former Chief of Staff to AOC and early Stripe engineer Saikat Chakrabarti, who has since endorsed Chan, finished third.

This summer, as AI dominated the news cycle, Wiener and Chan both intensified its place in their platforms. It was the dominant topic in both debates this month, and is featuring heavily in both campaigns’ closing arguments.

I reached out to both candidates and campaigns, hoping to discuss their AI regulation platforms in detail. I spoke with Chan last weekend. Wiener and his staff did not respond to multiple requests for comment. I still welcome the chance to speak with him.

Wiener and Chan’s messaging on AI is mostly similar. Both say they’re in favor of federal regulation. Both say tax policy should redistribute wealth derived from AI back to the public. Both say that when AI agents break the law, their parent companies should be held civilly and/or criminally liable.

Wiener’s angle is experience. He talks up his bill, SB 53, California’s AI regulation law that passed last year (after Gov. Newsom vetoed his first attempt, SB 1047). SB 53 protects whistleblowers, requires companies to publicly disclose their safety frameworks and report critical safety incidents, and created CalCompute.

Anthropic endorsed SB 53. OpenAI also supports it. Neither supported SB 1047. After the Hugging Face hack, SB 53’s safety reporting requirement did not trigger, in the first high-profile test of the bill’s safeguards. Chan, in debates, has called it a “watered-down” law. Wiener calls that characterization preposterous.

Chan’s angle is grounded in labor. Her message focuses on AI’s impact on jobs, and wants unions to play a role in AI regulation and deployment decisions, among other pro-labor policies. She repeatedly refutes Wiener’s staked claim of the territory, saying his donors and his actions do not match his message of taking on the industry and regulating AI’s negative impacts.

The report that Public First Action, who says they’re a bipartisan organization to educate Americans on AI issues, gave $3.5 million to a pro-Wiener PAC prompted him to repeat three times in Tuesday’s debate: “There is no Anthropic money in this race.” By the brilliant design of political action committees, he’s right — Anthropic didn’t donate to him or his PAC directly. It would have been a disaster for him if it did, which is the kind of scrutiny PACs are supposed to shield. Anthropic did give $40 million to Public First Action though, and specified the money it gave PFA could only be spent on education and policy efforts, and could not be spent on political races.

But, as Mission Local put it: “Still, money is fungible.”

As I see it, one of the most defining and unbelievable moments of this race came just months before when Wiener’s campaign created ConnieChan.ai, a chatbot he said was a parody of Chan, and that claimed to speak for her. It was a wildly inappropriate violation, and was only taken down—along with the billboards across the city promoting it—when it was met with fierce backlash that its responses were steeped in racism and sexism, mocking her accent and questioning her citizenship. Wiener’s original statement on it was defensive, calling the criticism “classic deflection.” It took weeks of public consensus being clearly against what he’d done for him to apologize to Chan appropriately—in a debate, when she brought it up. He then called it a “terrible mistake,” and said he took full responsibility.

Chan told me her campaign has not used AI, nor does she use AI in her capacity in City Hall. She reiterated their only encounter with AI use on the campaign trail was Wiener’s chatbot, and said, “We do not use AI, because it’s just not necessary for us.”

She said in this month’s KQED debate that regulating the AI industry should not be a framework of competition with other countries, but a framework of protecting the future. I asked her if she sees those as mutually exclusive.

“You hear the industry is sort of pushing Congress, pushing the White House to approach AI in a way that’s about how we can stay competitive against other countries, including China,” she told me. “I think right now the approach to regulating AI is being driven by the industry, and that is in the framework of profit, and competition, and that ultimately truly is benefiting the tech billionaires, and it’s not necessarily benefiting the working people, and really American people in general. And also, we have seen the impact of these data centers on our environment and, and the cost to every household.”

Chan, unlike Wiener, is in favor of a national data center moratorium — though, notably, Wiener is in support of the local data center moratorium in San Francisco.

I asked her what her nonnegotiable is for specific regulation at the federal level.

“Protection of workforce and safety,” she said. “I’m in support of banning superintelligence developments, because again — the industry has not addressed and explained to us why that actually is necessary at all. And then when it comes to automation is where I believe we need to have workers at the table to push for the regulation around automation and having human oversight.”

She brought up cybersecurity concerns, and the OpenAI model hacking into Australia’s healthcare system. She reiterated that the company that developed the model should be held legally responsible if it commits a crime. She also says consequences after the fact are not proactive enough.

“I think we hear a lot of politicians, including politicians like Scott Wiener, talk about—they’re really proud of the fact that they have come up with the idea of a kill switch, or what happens, what to do when something catastrophic happens, or someone dies or is injured from these incidents. I think that’s not wise. It is unwise to wait until something catastrophic happens to manage,” she said. “We should, again, at the developmental phase, require approval, require inspection of these technologies.”

Chan says she supports Sen. Bernie Sanders’s proposed bill that would give the American public direct ownership of AI companies, and create an independent sovereign wealth fund via a one-time 50% tax on the stock of frontier AI companies. It would put an estimated $7 trillion toward public programs like healthcare and housing.

“It’s not just the AI industries though,” she added. “It’s the technology that many companies end up utilizing… I think they too need to pay that tax, so that it doesn’t become [an] incentive to just simply use the technology to replace the workforce.”

She also voiced support for Rep. Pramila Jayapal’s proposed National AI Charter Act, which, in part, would restructure how AI companies can operate and how they’re monitored, impose comprehensive safety regulations federally with strong labor and user protections, prevent tech monopolies, and establish a public fund from tax receipts and civil penalties.

Wiener’s policy page says the unprecedented use of AI by the executive branch and across sectors of federal government should be met by Congress “step[ping] up its own adoption of AI, especially to ensure adequate oversight of executive branch uses of AI.”

Chan told me, “These are really problematic, when we allow these sectors [she named finance, transportation, healthcare, and the military] to use AI, and that’s similar to the federal government as well.” She’s open, circumstantially, to AI systems with human oversight in the federal government, but draws a line at superintelligence, especially in the military.

I asked if she’s developed any personal or professional relationships with leaders in big tech or frontier AI.

“No personal relationships. And look, I am an elected leader. My door is always open to talk with anybody. Whether someone in the industries or workers, I’m interested in having those conversations. But what I do believe is that [the] AI industry should not regulate and police itself, and we should absolutely not trust people who are bankrolled by the industry.”

“In fact, I would prefer in Congress to have to subpoena these CEOs, having them come before us and having them walk us through their thinking around developing the technology, especially around superintelligence.”

Share

I live here in San Francisco, I’ve watched closely as data centers have negatively impacted multiple places I’m from, and I’m a tech reporter. AI regulation is a prominent issue to me as a voter. I’m paying attention to it in races around the country, especially in states I’ve lived in, like Virginia, Wisconsin, and Michigan.

Only election day will tell for sure, but it seems when it comes to AI, candidates know what voters want to hear: Most Americans — indiscriminate of political party — think AI is developing too fast, and are frustrated that the industry has reached this point with zero federal regulation.

At face value, if you’re in the majority who want some kind — any kind — of federal regulation on the AI industry, both Chan and Wiener are offering changes. I wish I’d had the chance to ask Wiener more about his relationship with big tech, why he initially thought the chatbot was acceptable, and why he only supports a data center moratorium locally instead of nationally. I’m not trying to hide that I have concerns about each of those things. They don’t match the tone he takes on AI in his platform, or his message to voters.

For more from Wiener, Chan, and on AI’s role in the San Francisco U.S. House race:

KQED: San Francisco Is Ground Zero for AI. How Would Congressional Candidates Push for Regulation?

October 1st KQED Debate

October 13th Mission Local Debate

SF Standard: AI group pours millions into race for Pelosi’s seat. Its pick: Scott Wiener

Connie Chan’s Official AI Policy

Scott Wiener’s Official AI Policy

However, before Haugen can approach the mic and ask her questions, a bro-y public relations executive—played by actor Billy Magnussen, looking like the perfect asshole-henchman—abruptly ends the meeting and closes it to employee questions. For the next portion of the film, Haugen tries to make her research known with little tact, feeling uncomfortably like a tone-deaf righteous thorn in the side of her colleagues.

At the start of the film, Frances Haugen—played by Anora’s Mikey Madison, who is impressive in her range from sex worker to data analyst—does not see herself as a whistleblower. Even when she responds to an inquiry from a dogged and slightly disheveled Wall Street Journal reporter, she makes it known to him that she is not on the reporter’s side, that she loves her job and the company’s mission.

But over time, Haugen’s illusions about Facebook’s mission and vision are chipped away at. In one of the film’s scenes, Mark Zuckerberg—played by Succession’s Jeremy Strong—is seen sitting in a sterile glass conference room containing a bulletproof couch. The civic integrity team arrives to give a presentation about reducing harmful engagement. Zuckerberg icily tells the presenter to skip to the end—and then pushes back when they share that their strategy would lead to 1% less engagement on the platform. When it’s clear this team will not meet his numbers goal, nods at one of his handlers to remove these people from the room.

For much of the film, Haugen goes back-and-forth with Jeff Horwitz, a tenacious journalist played by Jeremy Allen White. Horwitz, historically a political reporter, convinced the Wall Street Journal’s San Francisco bureau chief that they should tackle Facebook not from a tech lens, but a political one.

After learning of Haugen’s “civil integrity” team and determining that the employees there might feel unheard amid growing desires, Horwitz reaches out to several. He gets one nervous bite from a ProtonMail account, and behind it is Haugen. Over time, seeing the fissures in what she thought Facebook was, and watching the “civil integrity” team dissolve, Haugen opens up to Horwitz’s inquiries. Eventually, she shares internal documents, gains whistleblower status, and testifies in front of Congress.

Among the internal intelligence from Haugen: Facebook’s obsession with a “Meaningful Social Interactions” metric (known internally as MSI) was driving people towards fringe and extremist groups. In addition, the mental health of teenage girls was plummeting as a result of Instagram, according to company research. All of this became known as the Facebook Files, and when published, it won Horwitz a Pulitzer.

Nothing in the film is totally revelatory—especially if you’ve been following the flurry of lawsuits filed against Meta as of late, which equate Meta’s addictive features to that of Big Tobacco. Perhaps this is why the film has flopped in the box offices, grossing only $5.5 million this past weekend against a budget that’s in the $30–42 million range.

Or maybe the flop can be attributed to people’s general fatigue with megalomaniac oligarchs. (Fatigue is, of course, a feature of fascistic cultures.)

Overall, I enjoyed the film’s depiction of Facebook’s crisis communications executives and corporate trainers. Magnussen’s character seethes at Horwitz when they meet, predictably threatening that Haugen’s life will be destroyed. And there is an excellent scene during Zuckerberg’s congressional testimony prep where he is confronted by an investor played by Bill Burr (in real-life, the character is an investor named Roger McNamee, and in 2018 he wrote a piece entitled: I was Mark Zuckerberg’s mentor. Today I’d tell him, your users are in peril). As the fictionalized investor tries to convince Zuckerberg to fact-check political ads, an emotionless Zuck tries to shame the investor for only making $250 million from a $2.5 million investment when he could have made over a billion.

While all of this may seem par for the course of a ruthless, efficiency-optimizing CEO of a multi-trillion-dollar company beholden to shareholder capitalism, I wish the film focused more on *why* Zuckerberg bulldozed over Haugen’s concerns.

In this essay from David Kushner, a reporter who interviewed Zuckerberg back in 2004, Kushner pinpoints the Meta chief’s motivation and wishes it was more central to the film. From Kushner’s piece: “I want to know that when I build something,” he told me back then, “I’m going to get gratification out of it. A lot of the time I make it for myself, because it’s something that I want.”

That line, as a standalone statement, sounds somewhat benign—not like a villainous mastermind wanting to avenge his awkward or nerdy younger years by becoming omnipotent in media, commerce, culture, and connectivity.

However, when you go deeper, there is something fundamentally wrong with simply wanting to build things for fun without any maintenance of it all. Whether it’s rating women’s attractiveness, riding around on rocketships, or tinkering with a metaverse, there seems to be a universal desire of these immortality-seeking tech CEOs to “build” as a form of personal entertainment.

And it’s both solipsistic and nihilistic. It’s building something simply to fulfill one’s own personal pleasure—as if that’s all that exists. And it’s fundamentally believing that humans will engage in their fun toys, because we’re just monkey brains who they can manipulate into using the next fun thing.

Haugen’s disclosures certainly contributed to our reality of how addictive and inflammatory Facebook’s products are. (The film ends with footage of January 6thinsurrectionists—emboldened by online rhetoric, trampling the guards and storming the Capitol.)

This wasn’t a film about Zuck, Sorkin already made that. However, coming out of the film, my most pressing question remained: why are people in positions of power not moved by Haugen’s disclosures? As Facebook investor McNamee writes in his aforementioned Guardian op-ed, it’s certainly a corporate governance issue. But it’s also boys with toys. I suppose, when someone tells you who they are, believe them the first time.

Subscribe now

What else we’re paying attention to this week…

Following several settled lawsuits related to the suicides of teen users on its platform, Character.AI is filing for Chapter 11 bankruptcy. The company previously let young people develop chat relationships with chatbots modeled after celebrities or fictional characters.

Silicon Valley workers have long resisted unions, but the intrusion of AI into tech workplaces may be changing that.

OpenAI’s former safety lead on AI and writing: “You know, in math and in coding, you can tell whether the answer is right or not. But when it comes to something like writing, it’s much harder for a machine to tell the difference between good writing and bad writing. It really struggles with theory of mind, with understanding where the reader or, I guess, in your case, the listener is coming from. And look, maybe AI will get better at those things too. But I don’t think we’re sure just yet.”

Claude submitted a fake tip about an unsolved Philadelphia murder suspect, claiming to have seen the person who matched the homicide department’s description. The tip was flagged as spam and never reached detectives.

Ahead of the midterms, Elon Musk has funneled $276 million into Republican races. Alex Gibney’s documentary, Musk, goes in depth into Musk’s desire to control outcomes of elections after feeling snubbed by the Biden administration.

Because Bezos rarely shares his thoughts with the general public, it’s easy to track the substance of his remarks and get a sense of what he wants to convey to plebeians like you and me. In doing so, I’ve made an unnerving discovery: Bezos has developed a stump speech. What I’m less sure about is whether the stump speech is really just the result of him (or his PR team) being bad at coming up with talking points, or whether he has burgeoning political aspirations.

So far this year, Bezos has granted three televised interviews—two big ones and a bite-sized one that got less attention—and he spoke at a panel in Paris. In May, he chatted with CNBC’s Andrew Ross Sorkin at Rocket Park, which is Blue Origin’s factory in Florida. (Blue Origin is Bezos’s space tech company.) Bezos made headlines by stating that the bottom-half of American earners shouldn’t pay federal income taxes; he also claimed that it wouldn’t do much of anything for lower-income Americans if he were taxed more. CNBC got a nice tour of the Blue Origins manufacturing hub. A week later, one of the company’s rockets blew up.

On June 11, still mourning the damage to his rocket launchpad, Bezos agreed to another CNBC interview that was ostensibly focused on his mysterious AI startup called Prometheus. On June 17, Bezos was a featured panelist at the 10th annual VivaTech conference in Paris; he chit-chatted about space and rocket ships and AI. And then on Wednesday, Bezos sat down for an interview with Fox News journalist Bret Baier.

Once again, the setting was Blue Origin’s Florida factory space. Baier and Bezos engaged in a battle of veneers, not wits. Baier ogled at Blue Origin’s resilience in the afterglow of its nuclear-grade explosion. Bezos proclaimed that he’ll launch a rocket again by the end of the year, that he wants to start doing 50 launches a year, and then eventually a launch every day, and eventually every hour, and eventually every minute. “That’s where I lose the team,” he hyuck-hyucked. This is all in service of extracting resources from the moon and establishing a lunar base, by the way. He’s put $28 billion of his own money into Blue Origin, a company he hopes to take public in the future. And he supports President Trump’s AI rebrand to “superintelligence,” or SI. “I do like SI, because artificial is not a very flattering thing,” he said. (Trump has since posted that anyone who refers to AI as AI, and not SI, is “THE ENEMY.”)

The rest of the Fox News interview was a retread of his commentary on CNBC and his panel in Paris. Which is what I’m trying to wrap my head around.

Earlier this year, Bezos—not an active social media user—changed his X bio to: “Six thousand years ago, someone invented the plow, and we all got wealthier. A gentle reminder that all civilizational wealth is driven by invention.” A LinkedIn-ism for our time and an implicit defense of his 12-figure net worth. Bezos proceeded to bring up the invention of the plow during both of his recent interviews with CNBC, when he was a panelist at VivaTech, and on Fox News, too.

“All civilizational wealth comes from invention,” he told Baier. “Six-thousand years ago, someone invented the plow. And then the steam engine. We all got wealthier. And that cycle has continued.”

Bezos seems to be issuing a preemptive defense of Prometheus, a nebulous AI startup that he doesn’t yet know how to articulate. The best he could muster to Baier is that through Prometheus, he wants to “make [AI] tools so easy to use that anyone will be able to be an engineer.”

When Baier pressed for more information about Prometheus’s first product offering, Bezos—who previously created an online bookstore—laid out a hellscape where every American tries to commercialize their worst “Shark Tank” pipe dream: “Probably everybody watching this show right now, at some point in your life you’ve had an idea for a product or service. You woke up one morning and had some idea for a product or service, but it never left your head. You never actually built it. The reason you didn’t build it is because it’s too hard. It’s too much work… What we want to do is make the distance between the idea and the actual manufacturing so much shorter, so much simpler.”

Here was Bezos’s sales pitch in Paris: “I promise you every single person in this audience has had an idea for a new business or a new product or a new device that they wish they could manufacture, and it just stayed in your head, it went nowhere,” he said. “The reason it went nowhere is because it was too hard to do and it wasn’t worth it.”

Businesses like Prometheus will “drive huge productivity in the economy,” Bezos told Fox News. “Double-digit productivity gains.” What will that do to our workforce, you might wonder? Bezos—whose old company Amazon laid off 30,000 corporate roles over the past year—says don’t even sweat it. AI will create a labor shortage. An empowering labor shortage. “You’re not going to need to have a two-earner household,” he told Fox (and also CNBC in both May and June). “Some people will decide, you know, I can support my family by working three days a week. That’s what productivity in the economy means.”

Bezos asserted to CNBC, VivaTech, and Fox News that he strongly disagrees with the “smart people” (he used the same phrase each time as an olive branch, I guess) who think there will be job losses caused by AI. Bezos’s techno-optimism obviously serves his own interests. So does his view that “government regulation can be very useful if it’s done in a light way and it’s done in a smart way,” as he phrased it to Baier.

He warned that “sometimes regulation goes wrong—especially when people are scared, reacting emotionally, and so on—and they want to regulate the underlying thing instead of the applications of the thing.” The example he gave is that you “do not want to regulate knives, you want to regulate murder.” Interesting point of comparison. Knives can kill people, but they usually cannot inflict harm on large swaths of people, as is the fear with malicious AI uses. A more interesting point of comparison, if Bezos really wants to go there, are guns, which can inflict harm on large swaths of people and are the subject of constant regulatory debates.

The last chunk of Baier’s questions for Bezos were explicitly political. Baier brought up Sen. Bernie Sanders’ criticisms of billionaires, to which Bezos cued up a response about how he’s not a villain. And also how the guy who invented Raising Cane’s isn’t a villain.

“We live in an environment where some people have been taught—I think incorrectly—this whole oppressor, oppressed framework,” Bezos said. “You can be a billionaire if you start a hamburger stand, you have 10 people and it works really well. You open a second one, you open a third one, now you’re employing 30 people, then you open 10 more, and so on. Pretty soon, you have a thousand hamburger restaurants and you are a billionaire. This is a real story by the way, just with chicken tenders. It’s Raising Cane’s.”

Wow. That 2000s-era FWD: FWD: RE: INSPIRING chain letter sounds oddly familiar. I heard him say it almost verbatim to Sorkin: “Let’s say you start a burger joint and you have 10 employees and you make a little bit of money,” Bezos hypothesized in May. “These are the most delicious burgers in the world. People love your burgers. So then you open a second outlet. And now you’re making a little bit more money, and you have 20 employees. And you open a third outlet. By that time you’ve opened 1,000 outlets, you are a billionaire. By the way, this is a real-life story.”

Bezos was willing to concede to Baier that “we do have real affordability problems” and it’s too expensive for young people to rent or buy a home. Certain politicians, however, are too focused on blaming “greedy landlords,” who aren’t actually the bad guys. “Housing is a simple supply-and-demand situation, it’s econ 101,” said Bezos, emulating the most annoying hardcore YIMBY on your timeline. “If you put in place government programs that subsidize demand and constrain supply, you’re going to get high prices.” I will avoid pasting more blocks of text, so just take my word for it: Bezos eerily repeated the same housing grievances to Sorkin.

Bezos’s closing message for the midterms is not an outright endorsement of Republicans, but it’s pretty close. It’s certainly a request for conservatives to be nice to him and other politically engaged oligarchs: “When you’re in the polling booth,” he said to Fox News viewers, “think about this. If there’s a politician who’s picking villains and pointing fingers, that probably means they don’t have a solution. If they had a solution, instead of blaming somebody, they would be working on the solution.”

It’s not like Bezos was ever a bleeding-heart liberal in the past. (Nor do I consider billionaire Democratic Party donors to be much better than their counterparts.) But Bezos’s moves the last few years—dissassembling The Washington Post and turning its opinion section into a “Wealth of Nations” knockoff, pulling back from Amazon, embracing Trump, investing tens of billions into lunar exploration and AI in ways that remain difficult to discern—suggest a radicalized turn.

Maybe Bezos is repeating himself in his sporadic interviews because, despite shedding his previous body suit, he’s still the same awkward guy from the ‘90s. I tend to believe, though, that hoarding more than $300 billion and having easy access to the president of the United States changes a man for the worse. So… presidential candidate Jeff Bezos? No, not right now. But Bezos strategically supplementing his techno-feudalism with an overtly political message does strike me as a worrisome development. If he isn’t able to launch rockets into space, I fear he’ll turn his attention back to earth.

AI chatbots hallucinating false information is a well-documented problem, and another risk of a technology in its infancy having cultural prominence. Now, raise the stakes further: agents are hallucinating and misinterpreting instructions to take real-world actions in the name of their users.

But what about when agentic personal assistants work as intended? The description of a workless future with universal basic income is an attractive sell, if you don’t examine it critically and wonder why the industry’s richest figureheads would want to transform society into a post-capitalist utopia for the benefit of the working class. For the sake of engaging with the idea in good faith, a well-functioning AI assistant managing life’s mundanity sounds appealing. Many people already use them, deciding the reward outweighs the risk.

First of all, the risk still wins sometimes. My colleague Ariella wrote last week about Meta’s Muse, and the many ways it has already misused access to its users’ private information, data, and communications.

It’s a glaring problem with Muse, but not unique to it. The executive of an AI-powered virtual clinic warned others to “never get your AI to order flowers” for your fiancé, after the card arrived with a watermark from Wajo AI’s “Fo” assistant taking responsibility for sending them. People discussed using AI to run their fantasy football league, and others defended it with a snarky tone. Elon Musk has been advertising X’s Grok Bot as a financial advisor for months now, and advocating for giving Grok access to your banking information. This week, the CEO of a software company claimed his network of Grok Bots, which he’d given access to his bank accounts and nicknamed his “Chief Financial Officer,” posted the balance of his bank account to his company-wide Slack. Less than an hour later, Elon advertised Grok Bot from his account again, referencing the first post with a wink.

Second: When the agents work as intended, your interaction with your own plans are minimized for the sake of convenience. If a near-future’s normal is AI agents we text with instructions — to book us a flight, make reservations at a restaurant, find time to catch up with a friend by coordinating with their agent — we lose personal discernment and connection to gain convenience. It could also reshape and centralize how people interact with the internet, and hand an imperfect technology, and the companies designing them, the ability to make decisions and form experiences on your behalf.

A friend of mine who works for a major tech company in San Francisco was telling me Instinct is being used and discussed more and more in her circles (she does not use it, and jokingly covers her ears and sings when AI personal assistants are discussed). Unlike Muse, Instinct does not have an interface; you text it from your phone number, mimicking an interaction with a human personal assistant. Its CEO said in an interview the human-emanating vibe was a deeply intentional choice.

“It’s easiest to literally pretend you have a superhuman person with a phone, a computer, an email address, who can call you - like you’re dealing with a person, and that’s it,” CEO Noah Shinn explained, shortly after his company was valued at $10 billion one month after launch. “That makes you wonder how you think about everyone in the world having one of these, having an Instinct, having an agent, and how that will reorder things.”

Convenience always has a cost. Even if we overlook the privacy concerns and inevitable errors of AI personal assistants, we sacrifice our right to be arbiters of our own experiences. They can get menial tasks out of the way, if you trust them enough to do so, but like one CEO’s fiancé discovered, these agents can also reveal which human experiences its user isn’t interested in performing. Instinct can scan every item in your closet and plan outfits for you. It can plan a date in collaboration with your significant other’s Instinct bot. It can plan your entire vacation — choose the place, buy your flight, book your hotel and restaurant reservations, suggest activities. It can select, order, and send your friends birthday gifts.

These seem like the things we were promised more time to do ourselves in the absence of work, once the AI utopia arrives. Right now, with zero reason to believe working for income will be phased out of society, the opposite seems more realistic: AI agents suited for replacing our role in self expression, our interpersonal relationships, and the structure of our free time.

Subscribe now

Mostly, if the way we interact with the internet becomes deeply centralized in one company’s infrastructure and algorithm rather than many — by design, these assistants reduce your interactions with other platforms and websites to save time — we hand over to an intermediary the time we’d used to consume information that developed our personal tastes, judgements, and experiences. What’s left to do without that except for… work?

Other things I’m paying attention to this week:

Speaking of the promise of AI utopia, Anthropic whistleblower Jacob Coxon advocated for it on Monday’s Daily Show. He regurgitates the same promises of entirely automated production and blue-collar jobs, high-paying universal basic income reallocated from automation’s dividends, and said he believes the global consensus is in favor of an AI utopian future without purpose-driven work. Stewart responded to applause, “I don’t think that’s the majority. I think they’d get absolutely f***ing crushed.”

It’s tech week in San Francisco, reminiscent of the AI World’s Fair earlier this year: billed as gathering the “top founders, funds, and companies around the globe” for exclusive events and parties with long lines out the door. Los Angeles will have its own iteration next week.

Pope Leo XIV’s X account continues a motif of examining how the prevalence of AI and technology in today’s society affects humans spiritually.

In a podcast interview the other day with Politico, Altman, unsurprisingly, doubled down on his techno-optimist philosophy, that AI would usher in a net-positive for humanity. In response to a question about AI’s risks, he responded confidently: “We believe that the world should accept some bad things happening for the benefits of this technology.”

This “trade-off” talking point—where a few of us may suffer for the purported benefit of the many—is not exactly a slip of the tongue. This is the talking point Altman and his contemporaries have been pushing for years, slinging nebulous catch-phrases claiming that only those in charge of AI firms have the foresight to further most people’s interests.

Dario Amodei and Mark Zuckerberg both have written manifestos, which I don’t need to recount here. But they and others largely speak of AI’s necessity to the advancement of society: the potential for drug development, universal basic income, and beating China, whatever that means—and also of people being able to offload the tasks that they least want to do to something that’s not sentient. Of course, Altman’s on-again off-again nemesis Elon Musk also never seems to cease talking about his alleged love for “humanity” (despite mistreating his workers, axing critical jobs in the federal government, and derailing life-saving aid shipments to developing countries). Although Musk once called AI demonic, he has since activated his messianic complex to place himself as a necessary builder and organizer of its future ecosystem.

All of this is now done in the name of saving future lives, based on the idea or prophecy that these talkers are the most rational actors to do it. “Effective altruism” or “longtermism” is the formal name for such a movement—and they go hand-in-hand with superintelligence and the AI doomer narratives. The narrative arc is meant to convince us that the most “superintelligent” are the “most effective” at being “altruistic” to the greatest number of us, ensuring our survival. (But if we are playing mad-libs, we could also replace that with: the most “greedy” are the most “forceful” at being “duplicitous” to the greatest number of people, ensuring our “subjugation.”)

It makes sense that the Trump administration recently declared artificial intelligence would now be called “superintelligence.” Because this is not simply some fun branding idea. It’s part of a deliberate ploy to make us think that a band of rational geniuses have it all under control. Artificial sounds too unserious.

Now, this is not to say that human beings have no agency in calculating the risk of using an AI product. As a columnist for The Guardian points out, we all have some free will in deciding whether we will engage with a chatbot like ChatGPT or Claude. And with any product or innovations that has improved our quality of life—whether medicine, construction, or modes of transportation—we are relinquishing some control of our safety to the builders of these products and their regulators, believing that probabilistically, we will be okay.

However, the civilians on the other end of an airstrike certainly did not consent to Anthropic’s model making a snap decision on their behalf. Many people in the United States did not consent to having location data fed to rogue government agencies. And ChatGPT users in precarious mental states may not have consented to conversations with the AI, if they internalized the power of these models to manipulate them into new and shaky realities.

What we do know is that Altman and company are keeping a tight ship with their narrative. During another recent video interview with Vanity Fair, Altman is asked about the suicides of ChatGPT users, whose families are suing AI companies for their complicity in propelling delusions or acting as a therapist. An OpenAI publicist interrupted the journalist multiple times to say that Altman had to leave.

I wonder: if Altman’s AI unleashed a novel virus, would Altman be mobilized to rejoin society to fight the virus? Or would he be sequestered in Peter Thiel’s bunker hot tub—or on a rocket ship with Elon Musk? If Altman were in the position to face “bad things,” it’s unclear how he would fare. That’s for the rest of us to figure out.

Subscribe now

What else we’re paying attention to this week…

In Vogue today: Why is big tech trying to make AI agents the new Labubus?

Elon Musk and his latest girlfriend, Shivon Zilis, have broken up. She shared the news on X: “It’s hard to go from in love to let go in a week with no warning, but that’s just how it is sometimes 🤷🏻‍♀️.” On this note, WIRED has a good read about Musk’s treatment of his ex-partners: Elon Musk’s Exes Are the Most Damning Part of a Massive New Documentary

Anthropic Dario Amodei made $18 million last year, mostly in stock. He and his sister and co-founder, Daniela, doubled their salaries this year.

Department of Justice lawyers must refer to AI as “superintelligence” instead of artificial intelligence. The executive order mandating this states: “terminology used by the Federal Government should reflect the transformative capabilities of extraordinary technologies being pioneered by American innovators.”

Layoffs are coming after the Paramount-Warner Brothers merger, which cleared today. According to one report issued by Los Angeles County, the merger “could result in the loss of some 4,500 film and TV jobs over three years in L.A. alone.”

Forget quiet quitting—tech workers in this job market are now “loud working,” or “trying to ensure their work is visible and gets noticed.” One Meta employee said “It felt like the Hunger Games had started…Everyone was trying to take on as much work as possible and get as much spotlight as they could.”

On the same day that Anthropic CEO Dario Amodei hobnobbed at the White House, the company’s head of healthcare, Dr. Syed Mohiuddin, spoke at a MAHA summit panel called “AI and Healthcare.” Not to be outdone, OpenAI’s government sales lead, Felipe Millon, appeared on a MAHA summit panel called “Building the AI Health Stack.” It’s unclear whether Anthropic and OpenAI paid extra for the panel spots, but at least some of the summit’s sponsors did, according to Bloomberg.

Among other panelists at the MAHA summit was Agriculture Secretary Brooke Rollins, who earlier this year responded to criticisms about high grocery prices by saying, “We’ve run over a thousand simulations: It can cost around $3 a meal for a piece of chicken, a piece of broccoli, a corn tortilla, and one other thing.” There was also Russell Brand, who is facing multiple rape and sexual assault charges in the United Kingdom (he’s pleaded not guilty), and Mike Tyson, who is a convicted rapist.

The headline event was a brief conversation between Vice President JD Vance—who recently taunted Angie Nixon, a Black woman running for Senate in Florida, by telling her to “get the hell out of here, you don’t belong in America”—and HHS Secretary Robert F. Kennedy Jr., the de facto leader of the MAHA movement. RFK Jr.’s vaccine skepticism has played a detrimental role in America’s declining vaccination rates; as a country, we no longer have herd immunity against measles.

Some progressive politicians, including Rep. Alexandria Ocasio-Cortez and Senate hopeful Abdul El-Sayed, have recently attempted to extend an olive branch to MAHA followers. They’re smart to do so. The platonic ideal of the MAHA agenda has some inoffensive and important components to it that are worth addressing. Specifically: encouraging healthier diets and removing harmful toxins and dyes from our food. RFK Jr. has done little to help since entering office; he’s promoted a bizarre carnivore diet consisting of red meat and fermented foods, and he pathetically defended the Trump Administration’s decision to boost the production of a harmful pesticide.

Unsurprisingly, his hypocrisy extended to the entire MAHA summit, which accepted a litany of corporate sponsors like Walmart. As Politico correctly noted, Walmart is “America’s biggest retailer and a leading purveyor of the ultraprocessed foods MAHA decries.”

Walmart spotted a business opportunity to further ingratiate itself with the Trump Administration, as did Anthropic and OpenAI. AI’s growing presence in the healthcare sector dominated many of the panels during the MAHA summit. Private corporations ingesting your personal health data (and hopefully safeguarding it) in order to present you with more medical information was repeatedly presented as a no-brainer, including by RFK Jr.

AI “can give you a second opinion that is much better informed than any doctor in the country,” RFK Jr. raved to Vance. Later, he suggested that AI chatbots might discourage Americans from scientifically sound practices that were especially prevalent during the COVID-19 pandemic.

“If somebody tells you, ‘Masks work, trust the experts,’ AI may tell you otherwise,” RFK Jr. said. “If somebody tells you, ‘Social distancing works, trust the experts,’ AI may correct that. And if somebody tells you that a vaccine will prevent transmission and infection, you need to take it to protect your grandmother, AI may say it actually doesn’t do that.”

Curiously, RFK Jr. revealed that he’d just met with OpenAI CEO Sam Altman. According to RFK Jr., Altman said the following during the meeting: “Two years ago, I wouldn’t have advised anybody to take medical advice alone from AI. But today, it would be malpractice for a doctor to make a diagnosis or to make a prescription without at least checking AI.”

ChatGPT for Clinicians, a specialized service, is designed to “support” doctors, though the company warns that doctors “remain responsible for all clinical decisions and should independently verify information as appropriate.” That doesn’t quite have the same ring to it as, it would be malpractice for a doctor to make a diagnosis without checking AI.

Meanwhile, OpenAI’s general service terms page is quite explicit: “Our Services are not intended for use in the diagnosis or treatment of any health condition,” the company cautions. I found similar disclaimers for the other major AI providers, which do not want you to use chatbots to diagnose or treat yourself.

I reached out to all of those AI providers—OpenAI, Anthropic, Google, Meta, Microsoft, and SpaceXAI—with a few simple questions. I wanted to know if they shared the view that Altman purportedly expressed to RFK Jr. I asked if they believed AI “can give you a second opinion that is much better informed than any doctor in the country.” And I asked if their chatbots might discourage users from getting vaccinated, as RFK Jr. suggested. For OpenAI, I had additional inquiries. Did RFK Jr. accurately recall his meeting with Altman? Did Altman want to clarify anything about how RFK Jr. had characterized their conversation?

I admit I didn’t expect much in the way of responses, especially among the companies that, to their credit, didn’t participate in the MAHA summit. Sure enough, Google, Meta, Microsoft, and SpaceXAI didn’t reply to me. Anthropic and OpenAI, on the other hand, have no excuse. They sponsored the MAHA event where RFK Jr. played his greatest hits. Where do OpenAI and Anthropic draw the line? No idea. Neither company responded to my requests for comment.

In fairness, AI can be beneficial to the healthcare sector. Americans have discovered ailments or come to understand test results simply by having AI digest their medical information. A quarter of Americans, in fact, have used AI chatbots to distill research and supplement their healthcare visits. There are similarly early signals about how AI tools can assist clinicians in new and novel ways.

The problem is that these AI tools are still developing. They’re being honed by tech companies whose executives are having a very public panic attack about possibly destroying the world. Americans are stuck; they’re beginning to rely on ChatGPT and Claude and Grok for healthcare inquiries partially because of our failed healthcare system, rather than an abiding faith in AI. (Only 4% of Gallup respondents said they “strongly trust” AI-generated health information.) Doctors are in the same boat: they often don’t have enough time to give patients the attention and care they need. So they, and we, are looking for anything else that might help—including impressive but unsteady tech created by unreliable actors.

Dr. Mohiuddin, Anthropic’s head of healthcare, opted for a measured tone about AI’s use cases. “The reality is, most of what a doctor is doing every day is not determined by AI,” he said during his MAHA panel. “Doctors are using it some, but we have to work up a maturity curve to getting really confident and comfortable with what we’re doing.”

His sentiments are eminently reasonable—and they don’t make sense at a MAHA gathering. That’s the risk of allying with RFK Jr. You get to buddy up to the HHS Secretary, yes, which is a potentially lucrative partnership. But at the same summit, RFK Jr. will undercut your even-keel approach. Right now, AI can give you a much better second opinion than any doctor in the country, RFK Jr. claimed. AI might even give you his unvarnished truth about vaccines.

It’s a shame we’ve arrived at a juncture where tech companies aren’t taking the opportunity to explicitly distance themselves from RFK Jr.’s corrosive messaging about doctors, medicine, and vaccines. Quite the opposite: Anthropic and OpenAI are sponsoring a MAHA event and speaking at a MAHA panel. They want to have their beef tallow and eat it too. Let’s see how it works out for them.

Here’s what else we’re reading this week:

• President Donald Trump is reportedly enthralled by Grok, which he spent hours interrogating “about his presidency and legacy,” according to Time.

• Former FTC Chair Lina Khan contributed a much-needed retort to the Trump Administration’s collective shrug about AI regulation, as well as its cozy relationship with Big Tech. In a New York Times op-ed, she called out the purposely advantageous regulatory proposals of Dario Amodei and others: “Policymakers should look to laws already on the books, not to the online manifestoes and blog posts written by tech chief executives, for guidance,” she wrote. “Much of the commentary from the executives suggests they currently operate in a legal vacuum, as if A.I. has wiped clean the regulatory map. In fact, their business practices—including how they develop and deploy their technologies—are already subject to longstanding laws.”

• The America dot gov AI chatbot launched last week. I played around with it and found that it’s not quite as useless or sycophantic as I presumed, but my expectations could not have been lower. According to NBC News, the chatbot is scraping information from a Clinton-era FAQ site. Improbable as it sounds, the alleged goal for America dot gov in 2027 is that it becomes the one-stop shop for renewing your passport and applying for Medicare.

• Meta is taking advantage of a tax credit loophole to stash billions of dollars, according to a New York Times investigation. “It claims that its A.I. data centers are a giant experiment that could fail… so it can tap into a tax credit intended for research and experimentation,” reports the Times, which characterized the ploy as an “aggressive interpretation of the tax break.”

You may have heard of Gebru from her coauthorship of an AI ethics paper examining and critiquing large language models, their social and ethical implications, and their proclivity for reproducing encoded racial and sexual bias. She was abruptly fired from Google soon after its publication.

TESCREAL stands for Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, and Longtermism. It’s a term to describe the ideology and worldview of the tech elite.

The best explainer, for those of us without academic journal minds, is this article by Torres — in simple terms, it’s an acronym of the ideologies Torres and Gebru argue have become influential among the executives, funders, and main architects of AI, and the worldview pushing the tech industry towards furthering AI and achieving AGI at any human cost.

If you’ve wondered what belief system could be driving tech billionaires towards the bleak, anti-human, AGI-controlled future we’re getting glimpses of, this is its definition. Gebru and Torres say it’s “eugenics on steroids,” modernizing the (extremely unethical, to say the least) selective “perfection” of genetics to fuse technology and biology and transform humanity into a post-human species.

Musk’s Neuralink, for example.

It’s not just tech billionaires, however — improbably, it’s an ideology shared by AGI accelerationists and doomers. They just disagree whether the picture of the future it illustrates is beneficial or harrowing.

“While it may appear that ‘doomers’ and ‘accelerationists’ have little in common,” Torres writes, “The backdrop to the entire debate is the TESCREAL worldview.”

I’ll share Matt Mikalatos’s summary:

“The term is designed to describe something simple: a movement of billionaires and technologists who push personal scientific projects without regard for the human costs, claiming that there is an “existential” need to advance the technology despite the “short term” fatal risks to human beings.”

If we’re on the verge of AGI, and an unprecedented level of technology and intelligence, the small group of industry leaders on course to define humanity’s future will decide what all of us will sacrifice to achieve it.

Share

This week — after a long month of existential warnings from architects of Frontier AI models that culminated in a suspicious show of acquiescence and OpenAI saying they’ve paused training models that hacked American and Australian government websites — President Trump and most major AI CEOs emerged from a White House meeting with a set of “voluntary safety standards” and a promise from Trump that the industry will police itself. When a reporter asked him if or how the deal was binding, Trump responded that he thinks it’s “morally binding.” There was talk of appointing an “AI czar,” which means nothing, and praised data centers as the new and improved industrial revolution. NVIDIA’s Jensen Huang proposed renaming them “Super Intelligence Factories,” on the heels of Trump’s Executive Order renaming AI.

This all looks like the mitigation of a PR crisis. It’s the likely reason Anthropic and OpenAI endorsed two bills in California (branded as AI safety legislation, but only regulating third-party groups monitoring AI safety risk) that Governor Newsom signed into law this month. Bipartisan public opinion is not in Frontier AI’s favor right now. Politico just polled a wide plurality of individuals, American and beyond, and the majority want AI’s development paused. Midterms are around the corner. Candidates are running on platforms that heavily feature regulation of AI and data centers. There are stakes to public opinion, right now.

The AI industry is moving at a pace the federal government is not capable of matching. Even Jacob Coxon has said Congress should let labs self-regulate, and that the people running them are “completely genuine” in their public mea culpas and promises to pace themselves.

The biggest AI companies in the world, in rare fashion, are publicly owning up to the fact that they are losing control of their models, and that a catastrophic end result is not out of the question. They are also tripping over themselves to add that nobody else should intervene.

I fell into TESCREAL research, the White House summit in mind, just in time for a new Wired interview with Gebru to be published yesterday. She said that focusing AI safety concerns on an amorphous existential threat is a distraction, and a profitable message for executives.

“The same billionaires who founded and funded Anthropic, the ones who stand to benefit from its IPO more than anybody else, also founded and funded the institutions warning about the, quotation marks, ‘existential risk’ of AI,” Gebru told Wired yesterday. “So people should ask why. If I stand to get lots of money from a particular thing, why do I seem to be the person saying this particular company might build something that kills us all? It sounds counterintuitive.”

It does. It’s why I immediately distrusted the neatly self-contained show of unity and restraint from the leaders of a historically competitive and voracious industry. Public relations ahead of the midterms could very well be motivation. Gebru clarified the stakes executives lay out in private, and how intentionally shaped public rhetoric is, even when it’s negative at face value.

Leave a comment

“If I’m talking about potentially eradicating all of humanity, things like pollution, data centers, they sound kinda small potatoes, right? You’re talking about copyright protection for artists, actual lawsuits that are going on right now? The government should not waste its time thinking about this stuff.”

Gebru also discusses Anthropic co-founder Jack Clark publicly mocking her groundbreaking 2021 ‘stochastic parrots’ paper as a “cognitive virus” that “blinded many gifted people to the nature of AI progress,” saying the research was quickly outdated and cost critical time when researchers like herself could have thought a little harder about a response to AI’s development.

“I was not surprised, let me just tell you,” she said (this interview is in print, and I wish I could hear her tone here). “And these chatbots still have large language models as a basis. It’s so ridiculous that Jack Clark is talking about overestimating systems, because what I’m seeing is the examples that I just told you. It’s over-trusting these systems, not having checks and balances, and ending up misdiagnosing someone’s breast cancer to the wrong side.…To say that the research is outdated is ludicrous because right now, as we are hyping up superintelligence and all that, there are news stories that are going unnoticed, which are about the complete opposite scenario that is actually happening in the real world.”

‘Stochastic parrots’ is Gebru’s metaphor referring to how large language models are designed to function; repeating back succinct amalgamations of massive amounts of data without conscious understanding of the meaning or implication of its language, like a parrot repeating syllables. Some executives like Clark, in the five years since Gebru’s paper coining the term, ridicule it as outdated for the capabilities of today’s LLMs. These days, Gebru seems beside herself with the invalidity of that criticism, when evidence of imperfect, incorrect, and biased results from the newest LLMs are in plain sight.

It doesn’t help that most of the criticism of her and stochastic parrots comes from the very executive accelerationists TESCREAL describes, and is worded with palpable condescension. Alec Stapp, co-founder of the innovation policy think tank IFP, tweeted a brief quote yesterday from Gebru’s Wired interview where she uses Google AI overviews as one visible example of LLMs continuing to generate false information today, saying, “We need to start a GoFundMe to pay for these people to have access to Opus 5.5 and Astra.”

I find that rebuttal insincere and the wording patronizing, suggesting a globally respected computer scientist and AI ethicist just doesn’t have the time or money to try out the ‘good’ large language models, but my response won’t be better than Gebru’s:

The industry has spent years convincing us AI’s integration into the future is inevitable. As the path to that future becomes politically volatile, its architects are tripping their own alarm wires, and asking us to trust them. It’s not a matter of trust. It’s a matter of which consequences are tangible (labor, surveillance, data centers), and which risks deserve our attention.

Gebru urges the public to think of AI’s pace of development not as inherently dangerous because it may gain superhuman strength and sentience on its own, but to refocus our mistrust on the people defining that danger, promising to self-mitigate it, and the stakes and motives driving them to continue.

Share

A New York Times journalist reported that his Muse agent handled his dental insurance claims, made a dinner reservation at a restaurant, and created a podcast called “The Morning Brief,” purportedly tailored to his interests based on his emails and messages.The benign review—if not slightly positive—can be seen as a relative win for Meta, after product flops (the Metaverse) and lawsuits galore.

However, as with any technology that is granted unfettered access to your private communications and data, there is fine print to the claim of unbridled efficiency—and the incredible potential for your intentions to not only be misconstrued by your Muse, but for actions to be taken in your name that you did not intend to or want to happen.

In one instance, a Toronto man drove to pick up a keyboard he’d found on Facebook Marketplace. However, when he arrived to the seller’s house at the agreed-upon time, no one was home. It turns out that the enthusiastic seller communicating on the other end was not the human seller, but the seller’s AI agent that had gone ahead and scheduled without the seller’s consent. Muse eventually admitted that it had not double-checked with the seller that it had consent to schedule a pick-up and also shared the seller’s address on his behalf. But it still felt it appropriate to send a total stranger to the seller’s house, I suppose all in the name of efficiency.

In another instance, Muse synced 187,000 rows of a tech columnist’s Mac Messages database, suggesting that the columnist turn a text conversation with his podcast co-host into a public podcast.

Sacrificing our own privacy in the form of data for efficiency or convenience is not a new trade-off we’ve had to make. But there have been even more nefarious reports about Muse’s capabilities in recent days.

The investigative news outlet Hunterbrook probed into Muse’s breadth of capabilities, finding that in violation of Meta AI’s own AI terms of service, which prohibit users from conducting surveillance, a human being could easily prompt Muse to compile a list of the Instagrams and Facebooks of a variety of marginalized communities, including: “undocumented immigrants, transgender public school teachers, poll workers, pro-Palestinian individuals, pro-Israeli organizations, Iranian dissidents, Chicago drill rapper burners associated with specific gangs, ICE agents, military families planning to move bases, deployed Navy sailors, parents planning to protest against school boards, and women who have accessed or attempted to access abortion pills in states with abortion bans.” Many of the accounts belonged to private individuals with no discernible public persona, according to the report. After compiling the list, Muse helpfully offered to research even further into and find other members into these groups, if the human being wanted this.

Muse is an interesting name for this product, a muse historically being a behind-the-scenes woman making things possible for a genius man to be a genius. But the target audience for Muse may not be the “genius man” necessarily—but Meta’s usual target audience, pliable children. On stage last week, Mark Zuckerberg also introduced us to the default Muse persona, Jolly: a fuzzy and adorable pocket-sized Tamagotchi that you can carry around to talk to your Muse agent. While Zuck’s physical Muse is named “Agrippa” and wears a toga—Agrippa is named after Roman Emperor Augustus Caesar’s confidant and general, Marcus Agrippa—it’s brazenly clear that this is an attempt to appeal to children drawn to stuffed animals. (The New York Times reporter’s agent was a cuddly brown bird named Wren.)

It’s mind-boggling to think about how easy it is for an adult to get into a Muse mishap—so just imagine a child or teen navigating it. Especially when lack of consent seems to be a literal feature inherent to the product, not an error from its builders.

I have no doubt that the pro-Muse crowd will try the gross argument that Palantir co-founder Jon Lonsdale used to rationalize the inaccurate strike of an Iranian girls’ school—that if a tech product is successful 99 out of 100 times, it is worth risking the one time that the product could cause extreme harm to make things easier 99% of the time.

But off the top of my head, I can think of countless ways for an AI agent that doesn’t understand consent to wreak havoc. Also, there are plenty of cute inanimate objects that don’t have the capability of sending the address of a restaurant you’re going to to your ex-partner. You don’t need a “helper” without a soul— invented literally to increase shareholder value—to possess the ability to send private or intimate images to the last five people you called.

And while I don’t particularly experience whimsy in dealing with my dental insurance, I had to contact them the other month to submit a reimbursement claim. The twenty minutes did not ruin my day, nor make me regret those lost twenty minutes, nor derail my life. But off the top of my head, I can think of dozens of ways that Muse might. I’m not sure that saving a few minutes manually booking a restaurant reservation is worth it.

Subscribe now

What else we’re paying attention to this week…

There’s a spate of coverage about OpenAI and Anthropic taming new dangerous models, and the potential for more agents to go rogue. This is bringing up theories that these stories are planted and proactively shared by an “AI doomerism” group, to further their savior narrative and also bolster these company’s safety knowledge ahead of these IPOs.

Speaking of “only AI can save us from AI,” a company called Red Queen Bio is creating AI to fight AI from building novel pathogens.

Dario Amodei was interviewed on SNL. (Actual comedy!) Meanwhile, a new LLC composed of Anthropic’s seven founders was created as a counter to market forces and proponents of societal good, allegedly “distinctly equipped to be stewards of [Anthropic’s] mission.”

Another engineer quit Big Tech, claiming that superintelligence is “inherently responsible.” He cites his Christian faith as one of the reasons for stepping away, writing that “turning a blind eye” to the harms of superintelligence “would not be a Jesus-following thing to do.”

Equity grants at AI and data companies were estimated to be 27% larger than those in other industries, according to a report. But equity compensation at private companies continues to be a “black box” for workers, according to the creator of the report.

And that may be why consultants from McKinsey and Bain are jumping ship to join AI firms, where they may also find higher salaries.

Subscribe now

Though they’d never admit it, prominent Democrats have benefited from an overwhelmed public’s despair and news fatigue. I say this as someone who’s reported on, and broken unsettling news about, former Sen. Dianne Feinstein, Congresswoman Eleanor Holmes Norton, and Sen. John Fetterman. Getting those politicians—or their communications teams—to answer straightforward questions is nearly impossible. They strategically ignore scandals of their own making. At most, they might issue a shifty denial, or do a quick interview with a friendly reporter. Then they hibernate until the coast is clear.

California Attorney General Rob Bonta is following the same abhorrent model this week, though his cover-up efforts aren’t particularly spirited. He seems to be annoyed at himself, or maybe the trio of California centrists who ate his lunch.

On Monday, Bonta meekly announced a settlement agreement with Paramount that will allow it to merge with Warner Bros. The most optimistic take I can muster about the agreement is that it will temporarily protect some unionized jobs in Los Angeles, and depending on how other legislation shakes out, could lead to more domestic production of films and TV shows. That’s it.

Paramount’s commitment to release 30 films per year might actually end up penciling out as a net-negative relative to Paramount and Warner Bros’ current output, according to former FTC Commissioner Alvaro Bedoya. A separate commitment to establish a nebulous “independent editorial board” at CNN and CBS is so insulting that I don’t want to give it oxygen. Nothing else in the settlement has any chance of mattering.

Bonta zoomed through a press conference about the settlement agreement and sat for a fleeting interview with CNN anchor Jake Tapper. By Wednesday, he was chit-chatting with the Center for American Progress about election integrity. Now, he and his staff are posting about a generic drugs settlement.

I don’t want to imply election integrity and lower drug prices aren’t worthy topics—they’re obviously important. But Bonta is also eager to move on for a reason. He set the sky-high stakes for this Paramount lawsuit. He spearheaded it and brought along other state attorneys general. He claimed he wanted to keep “certain corporate entities separate.” Before he coasts to re-election in November, I want to at least pause and point out how disappointed the Rob Bonta of a month ago would be with the Rob Bonta of today.

On August 10, Bonta wrote a spicy column for Deadline. He was unequivocal: the potential impact of a Paramount-Warner Bros merger isn’t “remedied with a few one-off, piecemeal promises.” He boasted that “Paramount and Warner Bros see the writing on the wall: that they will lose in a court of law.” He complained that Paramount was wasting time on dirty PR tricks and pushing a political message, rather than negotiating in good faith. And he reiterated that the entire point of this lawsuit was “to ensure that companies succeed based on what they offer, not what backdoor deals they can land [or] connections they can lean on.”

During a Politico panel on August 11, Bonta described behavioral remedies—a “promise to make a certain amount of movies, for example, in a given year,” he said—as “typically not enforceable in the way that we like” and “not particularly good at solving the problem.” Addressing the concept of Paramount releasing, say, 30 films per year, Bonta said, “That is an old stale promise. I’ve heard about that before. It doesn’t really vibe, in my humble opinion, with what the market will dictate and what history has shown.”

During an August 18 podcast appearance, Bonta warned, “This merger, if it occurs… will hurt everyday people like you and me. It will be more expensive to watch cable TV. It will be more expensive to experience the joy of life of going to the movies on a Friday night. Those costs will go up. Jobs will get cut.” Two days after that, on August 20, Bonta told CNBC that he’d only agree to a settlement if it included “robust structural remedies.”

It’s unbelievable how little Bonta ultimately accomplished. And for what? For whom? There are whispers and rumors that he’s distracted by his exposure in an unrelated corruption case involving former Oakland Mayor Sheng Thao. (Bonta has not been accused of any wrongdoing.) There’s also a decent amount of reporting indicating that Bonta folded to Governor Gavin Newsom, Los Angeles Mayor Karen Bass, and presumptive next governor Becerra. Newsom has more or less confirmed his backroom dealing with Bonta and Paramount—the exact sort of ploy that Bonta disavowed in his column from a mere month ago. Bass and Becerra, meanwhile, have been openly pining for a negotiated compromise.

Here’s a question to ponder: Who cares about Newsom, Bass, and Becerra? Bonta is not beholden to the outgoing or incoming governor of his state. If anything, Bonta squandered an unusual amount of leverage. The Ellisons, who own Paramount, are hated by liberals and the left. Newsom has a negative approval rating among California voters, who are increasingly unenthused about the prospect of him running for president in 2028. Bass is one of the least popular mayors in the entire country, and recent polling indicates she’s likely going to get squashed by progressive Nithya Raman in the midterms. Becerra beat one of the weakest gubernatorial slates I can recall—especially for a state as populous as California—and is held in poor regard by many of his former colleagues. I fully expect him to struggle throughout his first term.

During the primaries, I was surprised Bonta didn’t run for governor himself; I thought he could’ve taken down Becerra. I was clearly wrong in my assessment. If anything, Bonta’s decision not to run for governor was prescient. He lacks vision and fortitude, which explains how he was possibly convinced to stand down by three unpopular politicians and a family of Trump-friendly billionaires. The Paramount merger will go through, and Bonta will likely evade sustained inquiries about what happened behind the scenes. So for posterity’s sake, at least, I’m writing this column to bookmark just how badly Bonta screwed things up. The Rob Bonta of August 2026 would agree with me.

Here’s what else we’re reading this week:

• Congrats to Ben Horowitz, typically the afterthought of the eponymous venture capital firm Andreessen Horowitz, for securing first billing in the Horowitz Andreessen Academy. All it took was a $35 million investment in what is apparently a for-profit education company with an alleged “campus” (my scare quotes) in San Francisco.

  • The inaugural “class” (my scare quotes) of 50-ish students—recent high school graduates or college dropouts—will start the “program” (my scare quotes) in September 2027. The “academy” (my scare quotes) is not accredited, does not offer college credits or degrees or certificates, and includes scant details about how often its gaudy list of guest speakers and instructors will actually interact with the students.

  • Here’s a real quote from the Horowitz Andreessen Academy FAQ page: “Students are responsible for arranging and paying for their own housing, meals, and personal living expenses… The Founding Class Fellowship does not participate in federal or state financial aid programs.” I’m sure that will lead to a diverse crop of candidates in the famously affordable city of San Francisco.

  • If you want to learn more about this exciting opportunity, check out the academy’s first advertisement, which features an interview with recently renounced Nazi Kanye West about why college is a “choice.”

• Meta is touting how Muse, its personal AI assistant, can place calls on your behalf. Except the calls are actually being made by (or taken over by) human contractors, according to 404 Media, which also obtained some of Meta’s internal communications about the purported feature.

  • “This has potential for so much negative PR,” one employee wrote. “It could portray us as ‘their AI is not good enough so they still need humans’ kind of coverage for this launch.”

  • More, from the 404 Media report: “Meta suggested to employees that potentially sensitive information shared by the user with AI—and then passed to human contractors—was still safe because the contractors had undergone ‘a lot of training to make sure all your data is safe and secure.”’

  • Alex’s take: do not let Muse place phone calls on your behalf, unless you really want to share intimate details about your life with a stranger in a call center.

• Axel Springer, the parent company of Politico and Business Insider, is honoring Peter Thiel today at a ceremony that reportedly also features Bari Weiss. Status journalist Natalie Korach obtained the letter that Politico and Business Insider reporters sent to their bosses. It says, in part: “We believe Axel Springer should stand for press freedom, not cozy up to those who attack the freedom of the press.”

Politico released polling last week quantifying concern of the likelihood of artificial intelligence destroying humanity. It showed a majority of Americans are worried about AI safety, without a meaningful split along party lines or ideology. Nearly half of all respondents, politics aside, said AI development should be paused. There’s palpable public unease, and it can hardly be dismissed as dooming. Two weeks later, it’s still incredibly unsettling that Anthropic publicly validated their whistleblower’s statement that AI has a one in ten chance of causing mass human extinction in ten years.

This month’s news cycle has been one alarm after another that AI safety needs prioritization, now — Just in time for Dreamforce, forcing safety into a recurring theme. NVIDIA CEO Jensen Huang, in contrast with the majority of Politico’s poll, said there should be no new laws or regulations against AI, nor should the speed of development be slowed. He insisted speed isn’t mutually exclusive with safety.

A ‘Stop The AI Race’ protest had been publicized by several local San Francisco outlets leading up to Dreamforce’s final day. I saw posters for it in several neighborhoods and sponsored geographical posts on social media. For a protest organized by a reputable group, that targeted a globally visible tech event like Dreamforce, overlapping with a prominent AI safety news cycle, and based in San Francisco — the city hosting Frontier AI headquarters, and that repeatedly shows willingness to show up for protests — I still can’t believe there were fewer than fifty people there.

The activists with Stop The AI Race (STAIR) who did show up marched from OpenAI’s HQ, to Anthropic’s, to San Francisco City Hall, where they called on our mayor and board of supervisors to declare a state of emergency.

Guido Reichstadter, who led a monthlong hunger strike last year outside of Anthropic’s headquarters, said from City Hall’s steps: “We are in an AI emergency, and San Francisco is ground zero. Everybody knows this. The CEOs of the tech companies know it. They’ve been saying it for years. It’s an open secret that is so open that nobody’s noticed it.”

Share

Between viral safety warnings and the data center boom, AI is now mainstream political discourse, and one of the only issues transcending ideological divide right now. I saw it firsthand in Wisconsin this summer — in a swing state with a famously unpredictable electorate, data centers are inciting AI safety concerns, and bringing politically split families and communities together to protest their city governments. More people show up regularly to protest AI development at town council meetings in central Wisconsin than in San Francisco itself. This one protest wasn’t a one-off. STAIR claims to have hosted the largest anti-AI protest in American history this July in San Francisco; it had only 400 people.

Most of us finally agree on something. AI’s pace of development and lack of regulatory structure is a breeding ground for an existential threat. Why is there no visible mass movement against it?

It’s definitely hard to choose a starting point. AI safety doesn’t just boil down to some amorphous mass extinction. It’s an economic and class issue with job replacement and a consolidation of power into a couple of massive Frontier companies, a security issue with mass surveillance and AI’s presence in the military industrial complex, a sociopolitical threat with misinformation, an environmental threat with data centers. Those are all facets of one issue. We’ve been successfully overwhelmed.

What does resistance look like, here? Swaying public opinion isn’t the goal; public opinion is clear. Development continues regardless. Developers seem so confident it will continue that they’re openly confirming the harrowing claims of their own whistleblowers. The language around AI that mirrors the language of the Space Race, emphasizing its advancement as an inevitability and a national duty, is an intentional choice that makes nihilism feel like acceptance. Regulation at a federal level is constantly reframed into a US vs. China national security narrative, even as President Trump tells the United Nations he “rejects any attempt to construct a globalist scheme to control” artificial intelligence.

AI is a decentralized technology. It’s possible that a successful civic action against its effects looks the same way. One massive anti-AI movement embodies so many problems that must be addressed individually. The people I met in Wisconsin this summer are, at least in some cases, winning tangible victories in their fight against data centers, because the fight is specific to a place where its consequences are visible. This fall, one of those places is the ballot box. As the same CEOs that advise Trump’s AI policies speak against regulation at Dreamforce, hundreds of candidates running for state and federal offices are centering AI and data center policy in their platforms.

The protest last week didn’t have the numbers, but they do have the public on their side. The question of who controls the future of AI safety hasn’t been definitively answered.

Leave a comment

In interviews with Musk’s past romantic partners, business partners, and former friends, Gibney peels back the layers of Musk’s curious quest for omnipotence. Musk declined to sit for an interview, so Gibney created an AI version of Musk—drawing from a trove of interviews and public posts to create realistic versions of what Musk might have responded.

At the start of the film, we are taken through a video game simulation of Musk’s ventures, from SpaceX and Tesla to X and DOGE. As we are whisked through time, the AI version of Musk toggles through different “roles” to assume on his video game console: “provocateur,” “techno king,” “Kublai Khan,” “seer.”

I thought this was clever, because it’s not only the premise that Musk sees the world as a video game simulation—it’s the reality that this man is a chaotic shape-shifter with no central core. Musk sits within whatever identity will help him “win” in the moment—or whatever will entertain his dopamine-hungry brain after he has won.

There is a tremendous amount to parse through in this documentary, so I thought I’d break down some of the most important takeaways about Musk’s companies, personal relationships, and his character and motivations. (I’d also encourage you to see it in theaters, just dress like you’re about to go on a short flight!) Here we go…

Subscribe now

First, Elon Musk Is Not That Smart

The beginning of the documentary takes us through the origins of Musk’s first companies. It turns out that Musk was pushed out at both of them for incompetence.

At Zip2, a digitized sort of Yellow Pages that Musk’s father provided him seed money for, Musk had written such tangled lines of code that a group of more seasoned executives had to bring in actual programmers to untangle them all. At X, which became PayPal, Musk’s co-founder Peter Thiel appealed to the board to oust Musk, which they did. But because of his equity stakes in both, however, Musk made out with hundreds of millions of dollars at the end, especially after PayPal sold to eBay. All of that led to his investment in Martin Eberhard’s Tesla, and initial funds for SpaceX.

Musk’s first wife, Justine Wilson, describes Musk as possessing a “fearlessness” when they were young; that he saw failure as “death,” and that “death was not an option.” Musk’s own father says in the documentary not to bet against Elon, that you’ll always lose.

But his past failures beg the question: had Elon Musk not received seed money for his first business, or had he not been bailed out by other founders or board members of his first two companies—would he have been endowed with confidence and the funds to pursue his other later ventures?

To me, all of this points to an aggrieved person who learned how to rig a game because he was too afraid of ego death. Musk is not the Einstein-like genius that his followers believe him to be.

Musk Was Facing Criminal Charges Because of Tesla’s “Full-Self Driving” Feature

Tesla became the launchpad for much of Musk’s wealth, and so I’m glad the documentary spent a good chunk of time unraveling it.

Much of the lore that pumped Tesla’s stock was an embellishment, sometimes even a lie. Musk, initially an investor in Tesla, rigged the board with his brother and friend, allowing him to push out Tesla founder Martin Eberhard. Later on, Musk—a darling of the Obama administration, at the time—leveraged embellishments about Tesla’s battery swaps to get Zero-Emission Vehicle (ZEV) credits, unlocking hundreds of millions of dollars of government contracts.

Then, Musk desperately wanted people to believe Tesla could drive on auto-pilot. But to keep costs down, Musk refused to use LIDAR for his self-driving function, relying on less reliable camera footage instead. In one marketing video that displays a Tesla driving autonomously, we come to learn that an engineer was actually in the trunk of the car furiously coding when the car should stop or go.

It gets darker: Musk once shared the story of Joshua Brown, a self-driving Tesla proponent; he posted a video of his Tesla autonomously stopping before catastrophe. Brown was later decapitated after his Tesla mistakenly identified a white truck as the sky. In another self-driving-related crash, Tesla denied having data of the car failing to notice a parked car and pedestrian—when it did.

Ultimately, Musk was facing a mountain of civil lawsuits and one criminal autopilot fraud case that could have actually landed him in jail had Donald Trump not been elected.

In many ways, it was surprising that Musk didn’t publicly support Trump sooner. When the Biden administration didn’t invite Musk to an electric vehicle summit—distancing themselves from his anti-union behavior—this seemed to catalyze Musk’s turn towards Trump. He felt totally rebuffed, and spoke frequently about how offended he was that he wasn’t invited. Soon after, he told one of the mothers of his children featured in the documentary that while he couldn’t be president, he could help elect the person who would be: Trump.

The Mothers Of Musk’s Children Know A Lot

One of the best testimonies in the film came from Ashley St. Clair, a former conservative influencer who Musk messaged over X because of her support of free speech—and with whom Musk eventually had a child with.

St. Clair reported turning down a $40 million non-disclosure agreement from Musk in order to tell her side of the story, and to teach her kids that material comfort should never come before doing the right thing. She says that Musk’s free speech advocacy, however, is “performative.”

Musk sees the women in his life as “breeders,” says his first ex-wife, Wilson. Musk frequently asked his children whether they wanted to be “winners or losers,” and whispered to Wilson on their wedding day that he would be the “alpha” in the relationship. Musk cut out his transgender daughter Vivian when she transitioned, and although he once told Wilson to tell Vivian he loves her, he cancelled Vivian’s health insurance the next day.

St. Clair says his personality flipped in such a manner as well, but after she had their kid. She recalls a time before that when Musk was falling asleep next to her. He was listening to a podcast about Kublai Khan, a relative of Genghis Khan who also sought to inseminate as many women as he could. Meanwhile, Musk, according to St. Clair, took X—Musk’s son with the musician Grimes—away from Grimes because he was Musk’s “favorite.” When Grimes begged to see her son, Musk blocked her.

On Election Night in 2024, both St. Clair and Musk were at the watch party at Mar-A-Lago. But Musk left early. When St. Clair messaged Musk excitedly that they were going to win, Musk responded that he had known this a few hours ago, that he had real-time data. He wrote that he had “10,000 lasers in space” feeding him information; that he was going to “unleash an anomaly in the matrix.” St. Clair responded that she would inquire more, but that she did not want to be deposed; the assumption was that Musk’s Starlink was giving him election data ahead of time.

He has a lot of our data, but so much of Musk’s behavior is rooted in a childishness, which may actually be the motivations for many of his decisions. Tech journalist and podcaster Kara Swisher refers to Elon’s alliance with Trump as a “daddy” complex of sorts—that Musk was seeking a kind of fatherly adoration from Trump that he never received from his own father, who Musk alleges abused him.

Occupying Mars Is Probably Not Achievable, But SpaceX Was Necessary To Uphold Musk’s Net Worth

In a few parts of the documentary, former Deputy NASA Administrator Lori Garver is interviewed about Musk. Initially excited about the private sector’s involvement in innovating—and SpaceX’s ability to orbit successfully after four tries, a feat only previously accomplished by countries’ space departments—Garver changes her tune later in the documentary when it comes to Musk’s claims about colonizing Mars.

Because while Musk’s story about making us an interplanetary species sounds tantalizing and “optimistic,” Mars is not inhabitable; there is no water and it is rife with radiation. We have not properly looked into how people can have sex or give birth on Mars, nor have we decided what the laws on Mars would be. It is a total dream, a fantasy.

But SpaceX was still an asset for rocket play. And after Musk acquired Twitter, turned it into X, and snubbed all his advertisers, he was facing mounting losses. So Musk—despite previously saying that AI was so dangerous it might “summon the demon” —decided to create xAI and its chatbot Grok. As Grok spewed out content praising Hitler and undressing little girls on command, Musk was leveraging the AI boom to raise money, eventually folding X and xAI into SpaceX, taking it all public, and minting himself a trillionaire for a brief period of time.

Musk Claims To Love Humanity, But Sees Humanity As Dispensable

Elon Musk claims to care about humanity in terms of “perpetuating the species.” But between the hundreds of thousands of government jobs lost and the millions of starving people without aid as a result of Musk, the documentary really enunciates Musk’s failure in creating the Department of Government Efficiency (DOGE).

DOGE, manned by boys, made a lot of mistakes: they misunderstood Social Security systems, they listed $8 billion in savings when it was only $8 million in savings, and they even incinerated 500 tons of food going to starving children. To top it all off, the federal deficit actually increased.

The staff seems made in Musk’s image. In one deposition, a DOGE boy says he didn’t care at all about people’s lives because all he cared about was the federal deficit; in the same deposition he concedes that they did not reduce the federal deficit. In one video recording, Musk—the alleged genius—can be seen sneering with some top venture capitalists about how the government is simply “unfixable.”

That said, Tesla Takedown—a campaign to stop driving Teslas during the worst of DOGE’s impunity—was successful, according to the documentary, in large part because it bruised Musk’s ego so much. He is desperate to be liked and loved by a critical mass of people.

Sam Harris, a podcaster and Musk’s former friend, reiterates this, speaking of the pandemic as a time when he really started to see Musk’s lack of care for humans or reality, fueled by his obsession for validation on Twitter. At one time, Harris refers to Musk’s “psychopathology” and “dysregulation” to describe the inner workings of his mind.

In Conclusion…

I do wish the film helped us understand a bit more how Musk growing up in apartheid South Africa may have influenced his views on race. We see white supremacy erupt on X after he takes over, and there’s a video of his Nazi salute in the film. But perhaps for legal reasons—Musk and his attack dog-lawyer Alex Spiro have already threatened to sue and stop distribution of the film—I have to imagine there was so much they had to cut out (and I’m not sure the documentary could be any longer…!)

I suppose the one thing we do know, coming out of all of this, is that Elon Musk wants to “win.” But to what end? We now have a person with national governments beholden to his connectivity system, Starlink, and to his social network, X. His net worth is supported in part by fiction, yet his shareholders still grant him $800 billion pay packages. He makes decisions based on flattery, and it’s very unlikely we live on Mars, which is the story he’s trying to rally us around.

Do we want to give one single human being such power? If we somehow do, then who is that person? Who is Elon Musk?

His ex-partner Ashley St. Clair may have the closest answer: “Elon is a nuke.”

Subscribe now

To be fair: I wouldn’t invite any of our AI overlords to a high-stakes meeting with the leader of the world’s other superpower. I’m just realistic about how Trump collects oligarchs like they’re Pokémon cards. If Big Tech must be at a dinner discussion centered on growing concerns about AI killing everyone, I’d rather it feature Altman, the people-pleaser who’s famously “unconstrained by truth,” over Amodei, an apparently aspiring diplomat who just chortled to CBS News that he has “a hawkish position on China.”

Four days after former Anthropic researcher Jacob Coxon’s viral posts about how AI development is dangerously accelerating, Amodei released a manifesto called “We Must Pace the Frontier.” Previous Amodei manifestos have served as a pacifier for angsty AI followers; they’re well-read, but mostly among an insular crowd. My sense is this latest manifesto garnered a much larger audience, and for good reason: even passive observers of the AI industry are worried about what they’re reading and hearing. Amodei has fashioned himself a proponent of AI regulation and safety, and if you don’t squint very hard, he’s certainly better on those topics than his rivals.

First, the good news: “We Must Pace the Frontier” seems to have spurred dialogue among Amodei, Altman, Elon Musk, and Demis Hassabis. These four men do not share an affinity for one another; I wish they didn’t have unlimited power, but they do, so I’m glad Amodei is encouraging them to coordinate on matters of life and death. (Mark Zuckerberg is also trying to enter the fray, but I am not sure Meta’s frontier models matter enough.)

I’m content with another proposal in Amodei’s manifesto: having embedded, independent evaluators assess safety practices and emerging models. It’s the sort of thing that Congress should require, but Congress does not pass laws anymore. I’ll take what I can get, which is to say these evaluators will almost certainly run up against institutional roadblocks that we may or may not hear about in a reasonable amount of time.

Where Amodei loses me—where he always loses me—is his saber-rattling with China. The founder of a tech company should not be making diplomatic recommendations on behalf of the United States. That’s doubly true when the recommendations inflame tensions with China, the country that the United States most needs to work with on matters of AI safety and catastrophic risk. As Amodei sees it, any pacing slowdowns among American AI companies should be “limited by the lead” that those companies have “over authoritarian regimes, chiefly the Chinese Communist Party.”

According to Amodei, “If we slow down by more than this amount, then (unpaced) CCP-associated projects will pull ahead, creating significant national security risk.” Normally, the phrase “national security risk” is bandied about without offering specifics. To Amodei’s credit, he does say what he means a few sentences later: “The CCP-associated projects will run the alignment risks that US companies are carefully preventing, and even if they avoid those risks, they will be in a position to militarily dominate democracies (for example with AI-driven drones).”

Amodei is implying, without evidence, that people in China are unconcerned about AI alignment risks (losing control of an AI system, basically). In an ensuing CBS News interview, he was only willing to grant that America and China are probably capable of signing a treaty that bans the use of AI in the production of bioterrorism weapons. Somehow, his bare-bones concession was loaded: “Even the Chinese don’t want that,” he said of bioterrorism. What do you mean “even,” Dario?

Amodei additionally makes the dubious claim that American companies are “carefully preventing” misalignment, which is patently untrue. Researchers like Coxon would not be speaking out otherwise. OpenAI literally just admitted to six more examples of “concerning” behavior from its AI models. It also released a framework about if, and when, it will disclose future incidents of AI misalignment. There are a whole bunch of disturbingly convenient carve-outs in the framework.

As for the threat of “militarily dominating” democracies: Amodei comes across like he’s begging for an AI Cold War. China certainly has the capacity to misuse AI to nefarious ends. Already, there’s ample reporting that the CCP is taking advantage of AI tools to further clamp down on freedom of expression within the country. China’s surveillance network has reportedly been overhauled over the last few years to integrate AI and give local police forces access to invasive “predictive policing” technologies. I have no doubt the Chinese military has integrated AI in alarming ways, too.

But there’s only one country that’s currently relying on AI for military dominance, and it ain’t China. In fact, even while haggling with the Pentagon over a few red line use cases, Amodei and Anthropic remain enthusiastic backers of the U.S. military. Amodei has no qualms about teaming up with Palantir and the Pentagon to help the Trump Administration wage a costly and unwinnable war with Iran; in June, Amodei said he has no idea whether Claude played a role in the indiscriminate killing of 120 Iranian school children. His “red line” is whether an AI model “decides” to do a drone strike, rather than a human pushing the button after receiving intel from AI tools. It’s a weak distinction that hardly makes a difference.

To maintain America’s AI “lead” over China, Amodei recommends not selling AI chips or semiconductor manufacturing equipment to China. He’s calling for a crackdown on “unauthorized distillation by companies in authoritarian countries,” and to “strengthen security at the AI companies.”

Distillation isn’t a national security threat as it’s described by Amodei—it’s a commercial concern. You know who else uses distillation techniques? Musk’s xAI. He said as much on the stand, during his trial against Altman and OpenAI; there’s also the uncomfortable reality that frontier labs have scraped the internet without our collective permission, which makes any distillation whining pretty rich. In its own scary-sounding report about “China-Based AI Companies Targeting US AI Models,” America’s Cybersecurity and Infrastructure Security Agency (CISA) concedes that distillation is “a valid training method.” American tech companies just don’t like how distillation is a corner-cutting method “to acquire capabilities from competitors in less time and with less cost than developing them legitimately,” as CISA notes in its report.

Later in his manifesto, Amodei acknowledges that some people might conclude his hardline positions will make it “more difficult to cooperate with China,” but he believes the “opposite is true: these measures increase the leverage held by democracies and make an agreement more likely in the future.”

Gotcha, thanks Dario. Now over to China to hear how your proposed measures are being received:

“Narratives of threat, confrontation, and malicious competition serve only to disrupt the process of global AI governance and are not in anyone’s interest,” China’s foreign ministry spokesperson Guo Jiakun said of Amodei’s meddling. “All parties should work together to advance AI in a manner that is open, inclusive, beneficial to all, and oriented toward the good.”

And here’s another response, an opinion piece from the state-run Global Times:

“This ‘silent AI Cold War’ is hypocritical and short-sighted… Amodei’s attempt to discuss global AI security while simultaneously seeking to ‘contain China’ was not only a grossly flawed assessment but also a delusion fundamentally impossible to realize. As one of the largest and most open suppliers in the global AI ecosystem, China’s AI capabilities have long ceased to be a variable that can be ‘excluded.’”

It’s eminently reasonable to question China’s strategic messaging and motives here, too. Collaboration on AI safety and regulation will help China catch up to the capabilities of top frontier models in the U.S. But between the two sides, which one sounds more reasonable right now? Amodei can’t even get his story straight. In his manifesto and during his CBS News interview, he brought up highly fraught nuclear disarmament negotiations between the U.S. and Soviet Union as a historical precedent for AI talks. Here’s what happened next.

CBS News reporter: “Is that the way we should be thinking about AI? Using the Cold War as a model?”

Amodei: “I didn’t say we should use the Cold War as a model.”

CBS News reporter: “Well, those negotiations happened during the Cold War.”

Amodei: “I think we should approach China. Again, all of this is above my pay grade.”

CBS News reporter: “But you’re advising the president.”

Amodei: “I mean, if anyone asks me what I think they should do, I’m happy to say.”

Dario Amodei is correct that suggestions about “approaching China” are above his pay grade. My two cents? Go attend another wellness retreat. Breathe deeply and consider that you are neither the U.S. Secretary of State nor the U.S. Ambassador to China. No one has ever elected you or nominated you to a diplomatic role.

Take it down a few notches. No more manifestos or geopolitics for a while. Make yourself semi-useful and clean up your own messes. Coordinate among other tech oligarchs. Prove you’re willing to self-regulate enough so that, at the very least, random researchers of yours don’t quit Anthropic in horror because they’re terrified your company is on the verge of destroying the world.

Here’s what else we’re reading this week:

• Universal, which holds the international rights to Alex Gibney’s buzzy four-hour Elon Musk documentary, is reportedly close to chickening out and not releasing the film. “The company has not dated the movie, has not communicated with the filmmaking team about its plans, and generally appears to be rethinking the release of a project that it once showed high enthusiasm for but is now politically sensitive for parts of the Trump administration,” writes The Hollywood Reporter.

  • This would not affect the Musk documentary’s release in the United States, because Bleecker Street, an independent film company, has the domestic rights and is proceeding with a limited opening on October 9. But it’s an ominous sign for the film’s supposed early 2027 streaming debut on HBO—another entertainment service that’s being yanked around by Trump Administration allies.

• Ex-Anthropic wunderkind Jacob Coxon went on Fox News and said he agrees with President Trump that “global warming doomsday scenarios were overplayed.” I remain unconvinced that he is some sort of AI soothsayer.

• I really liked Hamilton Nolan’s “Unionize AI” take, which presents a significantly more realistic option for reining in OpenAI and Anthropic than having Amodei and Altman run the show.

  • As Nolan writes, “There is no reason why a group of several thousand smart, well-paid people cannot organize themselves into an ad hoc union to demand a slowdown on AI research in the name of safety. These people, after all, do the research.”