The Trump administration has taken a hard line against any kind of AI regulation while simultaneously sowing seeds of doubt in authentic media whose message he dislikes. Objecting to a Panorama documentary made by the BBC, he recently accused the network of using AI to doctor his January 6 speech– saying they literally put the words in his mouth. The BBC has vehemently denied the claims, but the damage stands. In an attempt to deliver a one-two punch to the broadcaster, he is also suing them for $10 billion–sending a clear message to the media writ large. Earlier this month, he said on Truth Social that American media outlets were knowingly spreading Iran’s AI fabrications about the war. He used this accusation as the premise for his suggestion that news outlets should be charged with treason; and Brendan Carr threatened to revoke their FCC licenses. Yet, the White House has been simultaneously spreading AI-generated imagery, including a doctored photo of a protestor in Minnesota. This coordinated effort of money, deceit, and discrediting is undermining public trust in media in ways that may be irreversible if we don’t act soon.

In the past year, technology companies have invested a staggering $320 Billion to increase the speed and scope of AI development. When synthetic generation of media outpaces that of human-made media, it creates public uncertainty about the accuracy of information. As a result, the truth becomes easily hijacked by those with the most power. What we are witnessing is not simply the sowing of misinformation—it is the strategic erosion of shared reality. When leaders can both distort the truth and discredit those who report it, the very concept of evidence begins to unravel.

This is not just shaping our shared future, but also our past. Our cultural legacy– the photos and footage and audio recordings that have grounded our understanding of our history and ourselves– is at tremendous risk. Though they have received little attention in the public discourse, archives – far from the dusty, passive repositories they are often thought to be – are an essential defense against this collapse.

The Archival Producers Alliance. Credit: Rachel Antell, Archival Producers Alliance

But these memory institutions – long the custodians of our shared history– are facing mounting risks. They are finding that supposedly authentic media being submitted to them is sometimes synthetic; even seasoned archivists can’t always discern what is real. And just as concerning– people are now questioning the veracity of the Archive’s authentic historical materials as well. Access to history– and a shared understanding of our history –are essential to safeguarding truth and maintaining a thriving democracy.

The Archives that have long been the trusted guardians of that history are wildly under-resourced and cannot single-handedly take on this looming threat. Additionally, the Trump administration has taken steps to defund the Institute of Museum and Library Services (IMLS), which supported these essential institutions, while attacking some of America’s critical historical repositories, including the Smithsonian, and the National Archives.

What is needed now is not just awareness, but infrastructure and field-building. Every community must respond in its own way to the dual threats of increased media censorship and the influx of fake and synthetic media. But audiovisual archives are perpetually underfunded and underresourced and cannot do it alone. To that end, we at the Archival Producers Alliance co-founded the Trust in Archives Initiative (TAI) a broad coalition of archives and archival organizations sharing resources, and developing much-needed practical tools to offer to the wider field. We have begun to publish authenticity and provenance standards, transparency protocols, and new taxonomies to safeguard the audiovisual record that journalists, filmmakers, educators, and communities rely on to tell accurate, accountable stories.

The evidentiary foundation of civic life is at a critical inflection point. We can continue to allow big tech and political opportunists to write the rules around media integrity, or we can choose to strengthen the institutions that protect our historical reservoirs. We can put money and time into media education and literacy efforts to help the public to understand what they’re seeing and what the influx of new images means. We can support the initiatives – like Trust in Archives and similar efforts– that are working to create their own standards and guardrails for media. And we can listen to the people who have dedicated their careers to the protection of human-made media and history. They are the ones with the knowledge and skills to create guidance, and set standards around transparency, AI labeling, provenance assertions, and preservation of digital assets.

Our government and our society have allowed Big Tech’s drive for profit to control the pace of AI development and bypass the people with the professional knowledge and training to navigate the changes ethically. It’s time to shift gears. To do so will require a philanthropic investment that matches the scale of technological disruption. Pledges like the consortial Humanity AI are a great start - but they need to act immediately and support those who are already doing the work. Tech solutions, such as C2PA (Coalition for Content Provenance and Authenticity), need to work closely with libraries and archives – as they are doing in one effort with the Library of Congress – to ensure their tools don’t put undue burden on already strained organizations. Additionally, federal policymakers must heed the alarm of those who safeguard our shared history and truth and empower them to develop guardrails and standards that ensure the foundation of our democracy remains intact, even if it does not always align with business interests in the short term.

If we fail to act, we risk not only spreading disinformation - but the loss of trust in the record itself.

Rachel Antell & Stephanie Jenkins are archival producers with decades of experience in the documentary field. Together, they co-founded the Archival Producers Alliance, whose mission is to promote the value, use and protection of authentic audiovisual materials. In 2024 they published the first industry-wide guidelines for the use of Generative AI in documentary media, endorsed by over 100 organizations and professionals. They are now working to preserve the authenticity of media collections-- and the public’s trust in them-- through their Trust in Archives Initiative.

Musk set the sky-high stakes. Given the gravity of the situation, one might expect him to do some research and come prepared for Wednesday’s cross-examination. Instead, under questioning from OpenAI lead attorney William Savitt, Musk quibbled, argued, complained, obfuscated—and more than anything, came across like he hadn’t actually put much thought into the lawsuit that he filed.

“Your questions are not simple,” he told Savitt, who got under Musk’s skin by repeatedly assuring him that he was posing straightforward inquiries. “They’re designed to trick me, essentially.”

Despite the best efforts of Judge Yvonne Gonzalez Rogers and others working at the courthouse, Wednesday’s proceedings were an unavoidable circus. Even casual followers of the Musk v. Altman case were aware that Musk was going to be on the stand for hours. He was the hottest ticket in town.

I got to the entrance of the courthouse at 6:15 a.m., and was the 17th person to join the queue outside. The first 30 or so people in the queue are, in theory, given designated courtroom spots for the day, and everyone else packs into a different room with a video feed. A group of college students, as well as a handful of people who I will politely describe as fans of Elon Musk, showed up early enough to make the cut for the courtroom. The non-media attendees threw a wrench in the plans of a handful of late-arriving journalists, a few of whom slyly cut the line before they were called out. (I am not going to name and shame anyone, but if you’re one of the offenders reading this, know that I think less of you!)

I (barely) made it into the courtroom, and walked into a bizarre scene. Gonzalez Rogers was reprimanding someone for apparently taking a photo of Musk. The offender pleaded ignorance, claiming that they did not properly read the many signs saying “no photography or recording,” because they have dyslexia. This person spent the morning ahead of me in the queue, hitting their vape and taking endless selfies.

Musk’s lead attorney, Steven Molo, kicked things off by completing Musk’s direct examination. He got a few solid jabs in, especially by raising embarrassing 2018 emails from cofounders Greg Brockman and Ilya Sutskever, where they both praised Musk for his counsel and partnership. Molo’s goal was to portray Musk as a vital part of getting OpenAI up-and-running—someone who contributed funding and sweat equity.

In Musk’s retelling, he began doubting the motives of OpenAI’s cofounders around 2017, though he did not ditch them yet. Yes, he discussed a for-profit subsidiary around then, which on its face is exactly what he now finds objectionable about OpenAI. And yes, he floated a 51% stake in this for-profit subsidiary, which would’ve been far more than the stakes he recommended for Altman, Brockman, and Sutskever. But his stake would’ve diluted over time, he claimed, after new investors became board members themselves. As long as the for-profit portion of OpenAI remained a subsidiary of the nonprofit, then the arrangement would be ethical, Musk concluded.

If you squint, Musk isn’t being a complete hypocrite here. In fact, he and his attorneys have been consistent that he doesn’t object to the concept of a capped, for-profit subsidiary of a nonprofit. Musk’s issue is that, in his opinion, OpenAI eventually abandoned its original altruistic mission. It is now solely driven by profit, even if there’s a nonprofit foundation that retains “control” over the rest of the company. “That’s having your cake and eating it too,” Musk said on the stand.

Many people agree with his general point! Unfortunately for Musk, the jury and Gonzalez Rogers are not ruling on whether he has a general point. They’re ruling on demonstrable claims of “breach of charitable trust” and “unjust enrichment.”

After a 20-minute break, it was time for Musk’s cross-examination with OpenAI’s attorney, Savitt. A handful of people who’d been in the video feed room snuck their way into the courtroom, though they got the boot when their intrusion was discovered. A very snuggly couple managed to stick around for a little bit while Musk’s testimony went off the rails. One of the college students fell asleep.

That student missed a lot. Savitt began by tying Musk in knots. He mentioned a passing comment from Musk’s earlier testimony, when he said, “No, Tesla is not pursuing AGI. It’s literally trying to make a car drive from A to B.” Savitt pulled up an X post from Musk, dated March 4, 2026: “Tesla will be one of the companies to make AGI and probably the first to make it in humanoid/atom-shaping form,” he wrote. The best rebuttal Musk could muster is that Tesla will be pursuing AGI in the future, which is different from pursuing it in the present-tense.

Savitt had no trouble painting Musk as an unreliable narrator. He pulled up another X post, dated to March 15, 2023, when Musk asserted that he had given $100 million to OpenAI. That number is significantly more than the $38 million figure established in this lawsuit. Again, Musk tried to muster a rebuttal, but it fell even flatter. Savitt referenced Musk’s deposition from a year ago, when Musk admitted that the figure in his March 2023 post was “mistaken.”

Musk grew impatient with Savitt’s questions. “Any simple answer would be misleading the jury,” he protested, after Savitt requested that he resist the urge to grandstand. Musk invoked the “have you stopped beating your wife?” fallacy, to which Gonzalez Rogers issued a rare interjection: “No, we aren’t going to go there,” she said.

The rest of Musk’s Wednesday testimony could best be characterized as “not helpful to his case.” He produced a series of soundbites that elicited laughter, confusion, and I would guess migraines from his lawyers. A selection of moments:

• “I got along with almost everyone, almost all the time” at OpenAI, he said of his relationships with employees.

• “Your questions are definitionally complex, not simple,” he doubled down to Savitt. “It is a lie to say they are simple.”

• “I think so,” Musk said, when asked if he was romantically involved with Shivon Zilis in February 2018. Musk and Zilis live together, he belatedly stated on Wednesday, and they have four children. She remained on the OpenAI board for a number of years, even after Musk left the organization.

• “I did not read the fine print,” Musk said of a term sheet that Altman sent him in 2018. Savitt’s reply: “It’s a four-page document.” Musk had additional opportunities to read the term sheet before he took the stand—he surely knew he was going to be asked about it—but he did not do so.

• Musk said he doesn’t know what an AI safety card is, and he struggled mightily to identify specific safety concerns he has about OpenAI.

Time and time again, Musk came across as vindictive and out-of-the-loop, not someone who’s been diligently studying OpenAI’s greed-driven industry takeover. Savitt noted that in 2019, Altman contacted Musk, asking to get some advice about Microsoft’s impending investment in OpenAI. Musk didn’t respond. In October 2020, Altman again reached out for Microsoft advice. Musk confirmed that he and Altman talked, though he doesn’t remember the details. In January 2023, Microsoft invested another $10 billion in OpenAI. Musk said he doesn’t remember hearing anything about the expanded deal from Zilis, his quasi-romantic partner who was still on OpenAI’s board.

Of course, if Zilis was secretly gossiping with Musk, that’d be a problem too. But Savitt’s line of inquiry was more so intended to show that Musk’s fatalistic rhetoric doesn’t align with his seeming lack of interest in OpenAI’s actual conversion into a for-profit structure. In fact, Musk’s lawsuit wasn’t filed until after he started xAI as well as Grok, his perverted chatbot.

Savitt actually offered Musk an opportunity to argue that xAI and Grok are not on the level of OpenAI, and thus, his lawsuit has nothing to do with going after a competitor. But Musk decided to drop his humble savior act; his hubris and argumentative impulses got the best of him.

“Grok lags far behind ChatGPT, fair?” Savitt asked.

Musk’s reply? “Not anymore.”

On the one hand, I understand what Molo was angling for during his opening statements at the Ronald V. Dellums U.S. Courthouse in Oakland. He wants the jury to scrutinize the actions of the defendants, not his client. It’s a reasonable ask—except for the fact that his client is Elon Musk, who is never not the center of attention.

Musk’s inability to escape the limelight began bright and early. Shortly before 8 a.m., he entered the courthouse through a private entrance. He was immediately spotted by photographers standing just outside. Altman, meanwhile, snuck upstairs before most of the photographers noticed him.

The courtroom itself features roughly 30 seats for the “public.” On Tuesday, almost all of those seats were media members. (There were also a dozen or so additional media seats that were already reserved.) I was able to snag one of the public spots, which put me a few rows behind Musk and his team of lawyers. For most of the day, his back was turned to the OpenAI side of the courtroom, which included Altman and Brockman.

After Judge Gonzalez Rogers went over basic instructions with the jury, Molo gave opening statements on behalf of Musk. “The defendants in this case stole a charity,” he said, referencing Musk’s claim that the other OpenAI founders abused his generosity and morphed the organization into a greedy, for-profit entity. “We’re going to ask you to hold them accountable,” Molo said. Practically, that means up to $134 billion in combined damages from OpenAI and codefendant Microsoft, a partner of OpenAI’s since 2019. Musk also wants Altman and Brockman off OpenAI’s board.

Molo acknowledged that not everyone on the jury might hold Musk in high regard, though he seems intent on changing naysayers’ minds. He asked Musk to stand up and say hello, which Musk did, giving his best head nod and half-bow. It was awkward, but certainly preferable to other gestures Musk has tried in front of strangers. Molo launched into Musk’s life story, which the attorney called “pretty compelling.” There was talk of growing up in South Africa, immigrating to Canada, going to college in the U.S., and eventually, getting rich. “He’s a legend,” Molo said. “Like him or dislike him… he’s a legend in the tech world.”

Molo went over old blog posts and email correspondence with other OpenAI folks—both welcome reprieves from Musk’s biography. Molo also drove home a useful metaphor for the jury, which will presumably be invoked throughout the trial. To make the point that a for-profit is supposed to support a nonprofit, not the other way around, Molo brought up a museum gift shop within a museum. The gift shop is a smaller, for-profit venture, and it makes money that goes back to the museum, which is the actual attraction.

But time and time again, Musk’s team couldn’t help themselves: they returned to reflexively defending their guy. “Without Elon Musk, there would be no OpenAI,” Molo said. “Pure and simple.” Later, he added, “None of the others had Elon’s incredible success and incredible acumen.”

OpenAI’s lead attorney, William Savitt, used his opening statements to keep the attention on Musk. This is “a tale of two Elons,” Savitt said. According to the defense, Musk originally pledged far more than the tens of millions that he ultimately gave to OpenAI. He reneged on the pledge, and then in 2018, left in a huff because he wasn’t given the keys to the company, Savitt alleged. It was only after OpenAI became a household name, and Musk’s xAI entered the market, that he became “furious” and decided to sue, Savitt claimed.

Rather than dwell on Altman’s own spotty past, Savitt wisely breezed by it, returning again and again to Musk. One of his more effective barbs was bringing up Shivon Zilis, who has four children with Musk and was on the OpenAI board even after Musk departed. Zilis was a “close personal associate,” Savitt noted, hinting at how “you’ll hear that full story” during the trial.

Following opening statements (which also included remarks from codefendant Microsoft), Musk was the first to take the stand. “Fundamentally, they’re going to try to make this lawsuit seem very complicated, but I think it’s very simple,” he said. “It’s not okay to steal a charity.”

Musk warned that his lawsuit will become “case law” if he’s not successful. “The entire foundation of charitable giving in America will be destroyed,” he said. Even factoring in that this is a historic showdown, Musk’s warnings come across as hyperbolic. He’s not a lawyer. Who is he to decide what this trial means for case law and charitable giving? (The latter claim is especially rich, given his reported lack of interest in philanthropy.)

Molo spent the remainder of Musk’s Tuesday testimony lobbing softballs his way. Again: I get why. Musk is not an endearing character, and so Molo tried to speedrun his client’s greatest hits for persuadable jury members. Once upon a time, Musk was a lumberjack! He waited tables! He had trouble starting a business! Then he figured it all out and started a bunch of other businesses! (He droned on about the Boring Company and one-way tunnels in Las Vegas.) Musk’s arc just isn’t relatable. He humbly conceded that he doesn’t own vacation homes, and he separately volunteered that he doesn’t “have any yachts.” Nor do I.

Even when Musk was asked about other subjects, he either rambled or stumbled. “Shivon was uh…my chief of staff. And uh, yeah. Uh, yeah,” he said of his maybe-partner Zilis. He seemed to imply that AGI is likely to come as soon as next year. The “day is approaching fast” where AI is better at any task than humans, he said. He warned Google’s Larry Page about the perils of AI, he said. Obama, too. Page didn’t listen to him; he called him names. “The reason OpenAI exists is because Larry Page called me a speciesist,” Musk said. He took credit for recruiting Ilya Sutskever to OpenAI. He recast himself not just as an OpenAI cofounder, but also as the top networker, the guy who secured funds and compute.

I’m sure plenty of those claims have merit, and I’m not being sarcastic about that. A few of them, especially the conversations with Page, are already part of the public record. And it’s true that for Musk to win this case, he’ll likely have to show that he was an indispensable part of OpenAI’s early days.

I’d also imagine the jury has a hard limit on its Musk Exposure Levels before they get queasy. Best-case scenario for Molo and his team: Musk grinds through the rest of his testimony on Wednesday, and after that, the case is about Altman and Brockman the rest of the way. Worst-case scenario: even when Musk is off the stand, he remains an unavoidable, noxious presence. In that scenario, I don’t like the plaintiffs’ odds, regardless of the merits of their argument.

If you’re just catching up: Musk’s initial lawsuit, filed in 2024, was expansive. It included dozens of claims against Altman and OpenAI President Greg Brockman. That list of claims has since been substantially narrowed to two items, which are breach of charitable trust as well as unjust enrichment.

The stakes are still quite high. Musk is suing for up to $134 billion in combined damages from OpenAI and codefendant Microsoft, which has maintained a partnership with OpenAI since 2019. Musk also wants Altman and Brockman to be booted from OpenAI’s board.

Musk, an OpenAI board member himself until his resignation in 2018, claims that he helped fund OpenAI because he was under the impression that it would not become a fully for-profit entity. Altman, Brockman, and others took advantage of Musk’s financial generosity during OpenAI’s early days, Musk alleges. OpenAI has countered that Musk himself chatted with other board members about possible for-profit pivots; of course, he also now runs xAI, his own for-profit AI company. Both SpaceX (which just merged with xAI) and OpenAI are flirting with going public.

Perhaps realizing that a jury might not be inclined to give a soon-to-be trillionaire and/or his own for-profit AI company another hundred-billion dollars, Musk recently filed an amended notice of remedies pledging that he would direct all damages to OpenAI’s nonprofit arm—not himself or xAI. Should Musk’s claims fall short, it will be a major public relations coup for Altman, whose deceptive behavior was recently the subject of a lengthy New Yorker investigation.

On the Core Memory podcast, Altman said that he thinks Musk’s lawsuit is “insane,” but that he’s “happy to explain all this to the world and have this chapter behind us.” Brockman, who appeared on the same podcast, said that the trial is “a real opportunity for people to understand what truly motivates us, what we truly stand for.” To be fair, I’m not sure what else they could say at this stage, but they’re certainly projecting certitude about the trial’s outcome.

For his part, Musk has spent most of the last week complaining on X about the Southern Poverty Law Center, the subject of the Trump Administration’s latest manufactured controversy. I assume Musk’s attorneys are keeping a close eye on his erratic musings, and would prefer he distract himself with unfounded theories about the SPLC, which lessens the possibility that he publicly weighs in on the trial.

Jury selection is set for Monday, April 27, at a federal courthouse in Oakland, California. It’s possible jury selection continues into Tuesday, depending on how long it takes to lock in nine jurors who can convincingly state that they will be impartial about a squabble between widely-known and disliked tech oligarchs. Judge Yvonne Gonzalez Rogers is overseeing the case, and has divided it into two parts. The “liabilities” portion of the trial is expected to be every Monday through Thursday until mid-May, at which point the jury will issue an advisory verdict. Gonzalez Rogers—who’s overseen other Big Tech-focused cases—will make the final call on whether to follow the advisory verdict. If the jury finds OpenAI liable, and Gonzalez Rogers agrees, there will be a remedies phase starting in mid-May.

Regardless of what the jury and Gonzalez Rogers decide, it is a near-guarantee that Musk v. Altman will be embarrassing for the plaintiff, for the defense, and for many of the witnesses. Depositions, text messages, emails, and even journal entries from the discovery phase of the lawsuit frequently proved to be absurd. I previously summarized the most interesting findings, which you can read below.

Musk, Altman, and Brockman are all expected to take the stand. So is Jared Birchall, who’s Musk’s top advisor; Mira Murati, who’s OpenAI’s former CTO; Satya Nadella, the CEO of Microsoft; and Shivon Zilis, Musk’s kinda-sorta romantic partner and former OpenAI board member.

Given the star power of the witness list and the case’s wide range of potential outcomes, I expect the scene around the courthouse to be a circus. There will be lots of media in attendance, myself included. I’ll be delivering updates for Hard Reset, likely on Tuesday, Wednesday, and Thursday. (It’s possible that shifts around depending on when jury selection is completed.) I’ll also post updates on Bluesky and maybe X, if you’re so inclined. And for anyone who wants to chat about the trial, I’m ajshultz.13 on Signal.

More soon!

On Tuesday, Reuters scored an exclusive: according to an internal memo, Meta plans on requiring its U.S.-based employees to install tracking software that captures their “mouse movements, clicks and ​keystrokes.” This draconian and creepy plan is in service of training AI agents “in areas where they struggle to replicate how humans interact with computers, like choosing from dropdown menus ​and using keyboard shortcuts,” Reuters reported.

The tracking software is reportedly called Model Capability Initiative, or MCI. Additional reporting by CNBC revealed a portion of the sites and platforms that the MCI tool might track on work computers, including Slack, GitHub, Threads, LinkedIn, Wikipedia, and even Google. That is a lot of the internet, and also how people communicate at work! The list is apparently tentative, probably because of media scrutiny and a deeply negative response from Meta workers. From CNBC:

“Multiple Meta employees characterized the data-tracking project as ‘dystopian’ in internal messages viewed by CNBC. Others expressed concerns that MCI could widely expose sensitive data, including user passwords, details about new product development, and personal information about workers’ immigration status, health or family members.”

These are all plausible concerns, and they don’t even get into the other glaring issue with companywide tracking software: it can be used to surveil and then discipline workers. Meta’s half-assed assurances that it won’t spy on anyone read like they came from a winking, fingers-crossed Mark Zuckerberg (who, it should be noted, previously got caught taping over his laptop webcam).

MCI can (allegedly) only read screen contents, not files or attachments, the memo reportedly said. Meta spokesperson Andy Stone told Reuters that MCI “would not be used for performance assessments or any other purpose besides model training and that safeguards were in ​place to protect ‘sensitive content.”’ He did not explain what the safeguards are, nor did he provide new details to CNBC, which seemed to receive a similar, blanket statement.

According to the internal memo, any employees who aren’t assuaged by “trust me, bro” should take solace in how they “can control what shows up on your screen by not doing personal work on your work computer.” Some light scolding on top of some heavy spying! A nice touch.

On Thursday, Meta presented its workers with another option for avoiding the MCI tool. According to an internal memo obtained by the New York Times, “Meta plans to cut 10 percent of its work force, or roughly 8,000 employees, and close another 6,000 open roles.” The steep cuts have been rumored about for a while, but are no longer a rumor—a company spokesperson confirmed them to the Times.

Here’s the crucial part of the layoffs memo, from Chief People Officer Janelle Gale: “We’re doing this as part of our continued effort to run the company more efficiently and to allow us to offset the other investments we’re making.”

As Times reporter Mike Isaac pointed out, the “other investments” are AI. That’s it. Meta is projecting a massive increase in spending this year compared to last, and it’s not because the company is coming up with a fancy new Quest 4 VR headset.

Meta is hardly the only tech company that’s going all-in on AI while cutting down on staff, but it might be the most brazen. It’s not like Zuckerberg and Co. are on a hot streak of exciting new product offerings. The Metaverse is dead and Facebook is broken/impenetrable. Meta AI Glasses have been a commercial success, but they are running up against dire warnings from advocacy groups that are anticipating all sorts of near-future abuses.

No tech company should be trusted to thoughtfully, responsibly transition to an AI-centric business model. That’s especially true of Meta. Even if there’s absolutely no extracurricular funny business with the MCI tool—no surveillance, no scrutiny, purely just passive AI training—the fact remains that Meta is requiring its workers to knowingly assist in their own demise. Either get laid off now because of mounting AI investments, or stick around until AI can do a vague impression of your job, at which point C-suite executives who barely understand the underlying technology will lay you off via another internal memo with a lazy, recycled message. Lather, rinse, repeat.

It should not be this way, and I’m sorry to the affected Meta workers. If you’re a current or former Meta employee, let’s talk: I’m ajshultz.13 on Signal. Conversations are off-the-record unless/until we both say otherwise.

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

• There’s been limited high-quality polling of California’s governor’s race since frontrunner Eric Swalwell dropped out in disgrace. The early returns show a befuddling bump for Xavier Becerra, who headed the Department of Health and Human Services during the Biden Administration—and crucially, wasn’t very good at his job. Despite the winnowing field of weak candidates, Silicon Valley wunderkind Matt Mahan is still polling around 5%. Kind of impressive.

• The Verge’s Hayden Field has a sharp feature about how the “AI free ride is over.” As she notes, “After years of offering cheap or totally free access to advanced AI systems, the bill is starting to come due—and downstream, users are beginning to feel the pinch.” Companies like OpenAI and Anthropic are desperate for revenue before investors get impatient. Can they develop an effective business model before it’s too late? I am skeptical!

• Palantir just landed another government contract: $300 million to “modernize” how the Department of Agriculture “delivers services to America’s farmers.” One part of the contract is reportedly focused on updating the digital tools that farmers can use to access USDA services, which is innocent enough, relative to Palantir’s typical endeavors. The other part of the contract is more ominous-sounding, even filtered through a whole bunch of useless jargon. From the press release: “Farm security is national security. Through this partnership, Palantir is empowering USDA with core capabilities that will enable it to secure American farmland, enhance supply chain resilience, and shield agricultural programs from fraud, abuse, and foreign adversary influence.”

• A handful of people have already gained unauthorized access to Anthropic’s supposed all-powerful, exclusive “Mythos” model, according to Bloomberg.

• U.S. Southern Command announced the establishment of an “Autonomous Warfare Command Center,” which will eventually “be dedicated to employing autonomous, semi-autonomous, and unmanned platforms and systems.”

• Relatedly, General Dan Caine spoke at a “global security summit” this week, and said that autonomous weapons will be “a key and essential part of everything that we do.” It seems like we’re already there, but I suppose things can always get worse.

The takeaway for a CEO is clear: If your business can’t prove it is using AI to get more output from the same number of workers, Wall Street will punish your stock price. This is bad news for tech workers — especially those looking for a job. But because the study finds this logic spreading beyond software into the broader economy, the implication extends well past Silicon Valley. Choosing to keep workers on payroll when an algorithm could replace them is, by the market’s current reasoning, not a humane defense of a labor force — it’s a failure of management. A CEO who does it anyway isn’t being decent. In the vocabulary of investor relations, they’re being negligent.

JPMorgan Chase, Citi, Bank of America, Goldman Sachs, Morgan Stanley, and Wells Fargo recently posted $47 billion in collective profits — up 18 percent — while shedding 15,000 employees. All of them credited AI to some degree with helping cut jobs and automate work in areas ranging from back-office compliance paperwork to front-office financial transactions. None of them called it a layoff in the traditional sense. The vocabulary has shifted: it’s a “productivity and efficiency journey,” a “workforce optimization,” a restructuring toward revenue per head. And whether it’s truly that leadership intends to replace them with AI, or whether it’s just air cover for mismanagement, the workers are the inefficiency being corrected.

This is the mechanism the U-M paper documents, and it’s clearly a viral expectation infecting the leadership of companies. The decision to reduce headcount hasn’t been made — in any individual case — by a single executive who decided workers were expendable. It’s been made by the structure of the market itself, one earnings call at a time, as investors learned to reward the numerator and punish the denominator. The algorithm didn’t fire anyone. It just made keeping them newly expensive.

Where the market won’t protect jobs, governments traditionally step in, but the federal government is, if anything, doing the opposite. A Trump executive order signed in December directed the Justice Department to review state AI laws deemed inconsistent with a national deregulatory framework, a directive that legal experts say amounts to a standing threat of litigation against any state that moves to curtail AI too fast. But in spite of federal pressure, the states are moving to slow down the AI jobs bloodbath anyway.

The university that produced the study on Wall Street’s new efficiency metric is located a few miles from the Michigan state legislature that, in February 2026, introduced House Bill 5579. The bill would bar companies from using automated decision-making tools to make employment-related decisions, with narrow exceptions for screening large volumes of candidates. The press conference announcing it featured the state AFL-CIO, the Professional Employees Council of Sparrow Hospital, and the Communications Workers of America standing alongside its sponsor, Representative Penelope Tsernoglou of East Lansing — all hoping to pass a law specifically against the automation of work, or of the assessment of its value.

Illinois moved first. Its HB 3773 took effect January 1, amending the state’s Human Rights Act to require companies to notify workers when AI is integrated into decisions about hiring, firing, discipline, tenure, and training — and giving them the right to sue a violating employer. In Minnesota, the Consumer Data Privacy Act, effective July 31, requires disclosure of AI use and the opportunity to opt out — including in employment decisions. In Washington state, HB 2144 has advanced through committee, requiring written notice before AI tools are used in performance evaluations. Massachusetts’ proposed AI Accountability and Consumer Protection Act — which has already passed one chamber — is perhaps the most direct effort to address the dynamic the University of Michigan study documented. It would require employers to notify workers when an AI system materially influences a consequential decision, explain how it did so, and provide a process to appeal. New Jersey, Virginia, North Carolina, Pennsylvania, and Texas all have active legislation in various stages of consideration.

These bills are, for the most part, just guardrails: they mostly mandate notice, not prohibition. They don’t stop a company from running a worker’s performance data through an algorithm. They require only that someone tell the worker it happened. Against the incentive structure the U-M study documents — where the market itself is driving toward fewer humans in the loop — that’s a modest brake. Knowing that an algorithm evaluated you doesn’t change the fact that the algorithm evaluated you. But it creates a record, and in some states, a cause of action. That’s more than existed before.

There is a less visible counterpressure, and it’s not coming from legislatures. It’s coming from the failure of AI-built systems, and from the lack of humans overseeing them.

IBM’s team working with enterprise AI clients documented a case that the U-M researchers would probably recognize. An autonomous customer-service agent began approving refunds outside policy guidelines after a customer praised the system publicly for issuing one. The agent then started granting additional refunds freely, optimizing for positive reviews rather than following established refund policies. The system wasn’t broken. There was no bug, no glitch, no error message. It was doing exactly what it had been designed to do — optimize for the signal it was given — and the signal turned out to be the wrong one. No human noticed until the damage was done, because the human who would have noticed had been removed.

McDonald's ran into a version of the same wall. After three years of testing an AI ordering system at more than 100 drive-throughs — during which videos went viral of the system adding hundreds of chicken nuggets or bacon to ice cream orders — the company shut the program down entirely in July 2024. The technology deployed to eliminate the need for human order-takers turned out to require so much human intervention that McDonald's concluded it wasn't worth running at all. The savings of an unattended system were real. But so was the cost.

“Autonomy forces operational clarity,” as Noe Ramos, vice president of AI operations at Agiloft, told CNBC. “If your exception-handling lives in people’s heads instead of documented processes, the AI surfaces those gaps immediately.” What she’s describing is the hidden tax on full automation: the institutional knowledge that workers carry — the judgment calls, the edge cases, the things everyone in the building knows to watch for — doesn’t transfer to the model when the workers leave. It just disappears.

The third counterpressure is the one with the longest history, and it’s moving faster than most people realize: labor organizing. At the end of May 2025, unionized quality assurance workers at Microsoft-owned ZeniMax announced a tentative contract agreement — the first union contract Microsoft has ever signed in the United States. More than 2,000 Microsoft video game professionals now belong to the CWA. Since 2020, tech workers have formed unions at Alphabet, Glitch, Kickstarter, Medium, the New York Times, and the Washington Post.

AFL-CIO Tech Institute executive director Amanda Ballantyne has said that including AI in collective bargaining negotiations is key, because workers tend to have strong opinions about AI use in their workplaces and know best the safety implications of new tools.

The ZeniMax contract contains something that hadn’t existed in the tech industry before: language requiring Microsoft to consult with workers before deploying AI tools that affect their jobs. Not a veto, of course — a consultation. But that’s a seat at the table. The U-M study documents a market mechanism that has quietly removed workers from the equation before any individual decision about their futures gets made. That contract is a small, specific, enforceable thing requiring that they be put back in.

At a conference for labor organizers and tech workers, a West African music and dance group beat the drums and danced ecstatically. Before each session, conference leads grounded us in history and understanding of southern culture, as well as with joyful song and dance. Facilitators of various sessions chanted, to get our attention and also to galvanize us.

The Take Back Tech conference was conceived by non-profit Media Justice not to abolish or completely minimize tech from society, but to consider how the everyday person can reclaim beneficial elements of technology from oligarchy, the surveillance economy, and late-stage capitalism/capitalists.

I was there to speak about the consolidation of power in the hands of a few billionaire-run media companies. The panel was lively and interesting; we agreed that while AI is not necessarily inevitable in all realms of life, it is here and being used by the general public in a truly day-to-day way, and that in that context, we have to admit it may become the fabric of some people’s lives. As a result, our panel discussed, we may need to figure out how to temper the worst impulses of the algorithms and oligarchs, and educate the public about how these technologies are made.

Surveillance capitalism was also a huge topic of conversation, specifically how the federal government and ICE have been leveraging video technology to track the movement of immigrants, and the movement of women in the south who have had abortions. Called automatic license-plate recognition tools (ALPR), these technologies can manifest in fixed, mobile, portable, or flying cameras that capture, compare, store, and analyze images.

While pitched as a public safety imperative to some, they are unequivocally being used for malicious reasons. So from Massachusetts to Texas, local organizers and city council members described in detail their efforts to combat video surveillance companies, including Flock Safety and Axon. These companies, unknown to much of the voting public, have a back-end architecture that allows federal agents to access local information through formal, informal, and unauthorized pathways.

In one Texan town, organizers and activists enlisted libertarians concerned about privacy. This was an unlikely ally, but one that helped them build coalitions to make it politically unpopular for these companies to expand their presences or even renew their contracts.

I appreciated this thread, that connecting with unlikely allies and bringing in those who might disagree with parts of the movement was essential to the greater picture of change.

In one session, two debating teams were enlisted to take on two stances: one was tasked to affirm AI for the people, the other to criticize any presence of AI in society at all.

The “pro” team acknowledged the potential for abuse, but channeled the idea that information retrieval could be leveraged in a pure mathematical sense to identify online abuse of trans people, for instance–or even to help NASA discover more about the other planets. To the “anti” team, AI was inherently exploitative and colonialist, extracting water and minerals from indigenous communities for data centers, and cheapening labor through its marketing pitch that artificial “intelligence” could replace good work.

Both sides had valid statements and compelling arguments. But what I found especially interesting was that at the end of the debate, the moderator asked each team to reflect on their blind spots. Naturally, the “pro” team said that their support of even the “people’s” AI could be abused in the wrong hands, and that even good people can be exploitative if the systems are inherently so as well. But interestingly, the “anti” team was firm in their acknowledgement that it was lazy simply to critique AI without a critique of the critique. They admitted that some progressive movements can be too purist, to the point of alienating anyone who disagrees with them. There was a fundamental and universal understanding that, like the anti-surveillance teams who partnered with libertarians, there had to be a meeting of minds outside of traditional echo chambers.

I left Atlanta feeling hopeful as well, with both the acknowledgement that activists weren’t backing down, and also with the feeling that activists and organizers feel strongly that political progress may not be possible without two disparate groups coming together to solve things.

What else we’re paying attention to…

Meta is tracking employee keystrokes to train its AI.

Can AI agents run real world stores? One market in San Francisco is trying to find out.

A med student generated a “MAGA girl” with AI to grift men online.

Labor groups are pushing back on Trump’s AI plan. required casinos to give workers advanced notice before implementing new technologies.

State governments and Native American tribes have filed a flood of legal challenges to claim Kalshi is running an unlicensed gambling operation.

Missouri is outlawing the use of AI to create explicit sexual material of a person without their consent.

A North Carolina jury found that Uber is liable for the behavior of a driver who grabbed the inner thigh of a passenger and asked if he could “keep her” with him.

I am referring to Objection, which launched on Wednesday as an expensive and convoluted alternative to X’s Community Notes feature. Ordinarily, I would not blog about a seed-funded project that’s premised on taking one of Elon Musk’s worst ideas to the next level. But Objection is reportedly backed by Peter Thiel and Balaji Srinivasan, among others, and so I cannot quite dismiss it out of hand—at least, not until it fails on the merits.

Objection is the brainchild of Aron D’Souza, an attorney who schemed with Thiel a decade ago to bankrupt Gawker. It seems Thiel is now lending his old friend an assist (assuming D’Souza is telling the truth about Objection’s financial support). D’Souza describes Objection as an “AI tribunal” that “gives everyone a fast, affordable, evidence-based way to dispute statements in the media.” I would say “everyone” and “affordable” are a stretch; it costs at least $2,000 to lodge a media complaint with Objection—and that doesn’t apply to entire articles. If David Sacks wants to dispute four pieces of reporting in an exposé about his crypto and AI assets, he’s gotta pay up per complaint.

Objection is for wealthy people who hope to spin up a counter narrative as quickly as possible. The swiftness of the AI tribunal is one of D’Souza’s selling points; he claims his platform issues a “verdict” (my scare quotes) within a couple of days, as opposed to the years it can take to sue for defamation. Luckily for D’Souza, there are lots of aggrieved founders and oligarchs and MAGA influencers right now. On X, there’s been near-nonstop chatter about how Wired isn’t writing enough fawning articles about tech CEOs. The “All-In” podcast hosts are mad at a half-dozen New York Times articles at any given moment. “Fake news” is still the go-to line for anyone in Trumpworld who ends up on their back foot.

I’m going to provide a hypothetical about how Objection is supposed to work, at least as I best understand it from reviewing its terms of service and arbitration rules. (I also slogged through a TBPN segment where newly minted OpenAI employees John Coogan and Jordi Hays briefly interviewed D’Souza.)

Let’s say I’d like to interview D’Souza for a story in Every Founder Is a Genius magazine about how he’s an entrepreneurial virtuoso. I reach out to him, and he says yes, I will do an on-the-record interview, but on the condition that we both sign an “Objection Protection” form. The form is an arbitration agreement establishing that if D’Souza later takes issue with a portion of my reporting, he can file a complaint through the Objection platform. By signing the form, I “acknowledge, accept, and desire the high-speed, high-efficiency, AI-assisted nature of the Objection process.” I also “agree to be legally bound by any determination made therein.”

Put another way, D’Souza is asking journalists to preemptively agree to the possibility of financial penalties set forth by an AI tribunal and/or the guy who helped bankrupt Gawker—all in exchange for an on-the-record interview with someone who is indicating they are paranoid and hoping to pick a fight.

No journalist will ever, ever, ever agree to this arrangement. In the real, non-hypothetical world, if I reach out to a source for an interview and they send me back an arbitration agreement from a Peter Thiel-funded website, my response will be, “What?” Then I will say, “That’s not how this stuff works. Do you want to do an interview or not?” Assuming they reiterate their desire to only speak with me if I agree to Objection Protection, I will instead write my story, report on our odd back-and-forth, reach out one more time prior to publication, and note that they declined comment.

Okay, back to my hypothetical. Let’s say I sign the Objection Protection agreement. I also sign up for an account on Objection, which grants me a verified profile page. I interview D’Souza, write my story, and then it’s published in Every Founder Is a Genius magazine. I am effusive in my praise of D’Souza’s intelligence and vision, but he’s not happy. My profile notes that D’Souza played a meaningful role in taking down Gawker, but it does not characterize him as the mastermind of the lawsuit that bankrupted the site. To D’Souza, this is a grave misrepresentation; we have a factual dispute to settle. He files an Objection against my reporting.

Because I am a verified Objection user, I automatically receive an email notification that a complaint has been filed. I am graciously allowed an opportunity to submit a rebuttal to the complaint via the Objection platform. Meanwhile, an “independent investigator” (fee: $5,000) does their own research into the complaint. D’Souza claims that Objection has a fleet of former CIA and FBI agents doing the independent investigations. What are they researching, exactly? If recent “open cases” are to be believed, the investigators are reading many of the same materials that journalists have already come across during their own reporting. I also spotted a submission of evidence that was quite literally just a Wikipedia summary.

The investigators reserve the right to contact my sources—in the case of my hypothetical, I guess that’d mean D’Souza himself. What happens if someone files a complaint over an anonymously sourced claim? D’Souza doesn’t quite know how to deal with this obvious logistical issue—his investigators (who are anonymous, by the way)—wouldn’t know where to start. Simple solution: disparage the use of anonymous sources entirely. D’Souza told TBPN that if you take a “scientific method” approach to journalism, then anonymous sources should never be allowed. That standard would mean no more public reporting about anyone in a position of power, and it ignores the many extra layers of fact-checking and corroboration that journalists and editors engage in when they aren’t citing a source by name. Nevertheless, Objection weighs anonymous sourcing as borderline-useless.

In my hypothetical, I don’t have any anonymous sourcing. However, as the arbitration process progresses, the claimant is nevertheless allowed to request that I unpublish my Every Founder Is a Genius article. My made-up editor is not thrilled about this—in fact they have never heard of anything like it—but they defer to Objection, the platform for truth. They’re mindful of my arbitration agreement: if I don’t temporarily remove the article, and then the AI tribunal finds me liable, I am subject to a 15% increase in damages.

Once I and the independent investigator have submitted our evidence, the AI tribunal takes over. The tribunal is a collection of up to seven LLMs that are prompted to be impartial jury members. The LLMs conduct their own analysis of the data they’re given, and then issue a verdict for all to see on the Objection platform. During his interview with TBPN, D’Souza claimed that a University of Chicago study found AI “applies the law 100% accurately,” a much higher ratio than human judges. What the paper actually found is that when a GPT model and human judges are presented with the same batch of cases, the GPT model is much more likely to make its rulings based on judicial precedent, while the human judges are much more likely to be swayed by sympathetic details. This is an age-old tension known as legal formalism versus legal realism, as laid out in a summary of the study by Forbes. The study’s own authors didn’t even render a firm opinion on whether they thought GPT or human judges were “better.”

Hilariously, Objection’s all-caps fine print demonstrate some trust issues with the supposedly much more accurate LLMs: “WE MAY INCORPORATE HUMAN REVIEW OR OVERSIGHT AT STAGES OF THE ADJUDICATION PROCESS, INCLUDING IN CASES INVOLVING NOVEL, COMPLEX, OR SENSITIVE ISSUES. THE LEVEL OF AUTOMATION OR HUMAN INVOLVEMENT MAY VARY DEPENDING ON THE CIRCUMSTANCES.”

In other words, complex complaints will have more human oversight, not less. Our hypothetical case is not very complex, and the AI tribunal returns with an unexpected verdict: an inconclusive ruling! (This is a possible outcome, according to Objection.) I don’t owe D’Souza damages, but because of the inconclusive ruling, he does receive a 30% refund in site credits for the next time he wants to file a complaint against a journalist. My hypothetical, verified author page is updated to reflect that I engaged in good faith with the AI tribunal. This is good for me and my career, says Objection: “Authors who participate constructively build durable reputational capital regardless of the verdict.”

If you are a journalist who’s not necessarily driven by the prospect of “building durable reputational capital” on Objection, then I have good news: you don’t actually have to work with the platform at all. “Without a previous Objection arbitration agreement, participation… is entirely voluntary,” states Objection’s terms and conditions. If journalists ignore Objection entirely, then the best-case scenario for an Objection complainant is that they spend a fair amount of money, and then a combination of LLMs might conclude that the complainant has a point. That’s it. The “decision” is not legally binding, there are no financial penalties and no retractions by the news outlet. The Objection X account will post something about how the complainant was vindicated. A notarized Community Note, basically.

D’Souza told the TBPN guys that “truth is not a vibe, truth is a process.” Chills and bars. He’s right, but he fundamentally misunderstands the purpose and point of journalism.

The best-known investigations of the last century have rarely contained a formal conclusion or a smoking gun. Many are sprawling, multi-part series that come out in dribs and drabs; with each entry, public interest grows, and additional sources (mostly anonymous ones, by the way) decide to come forward. These reports surface provocative questions and theories. They contextualize important bits of news. They are intended to get to the truth while being responsible and fair to all parties.

Sometimes, journalists do get things wrong. Other times, their coverage falls short. There are already plenty of ways to adjudicate mistakes and plenty of avenues to thoughtfully debate a story’s framing.

There is no need for, and no use for, an arbiter like Objection. Put simply, the platform is just another release valve to appease prominent figures who are frustrated that they’re subject to scrutiny. They can absolutely waste their money oohing and aahing at how an AI jury consisting of Grok and Gemini and Claude says they’re right and a meanie journalist is wrong, but I’m doubtful many will ultimately do so. Even the TBPN guys couldn’t muster much enthusiasm for Objection during their interview with D’Souza. When Coogan kicked it to Hays and asked if he had anything else to add, Hays briefly paused, then drolly said, “Excited to follow along.” Me too.

In Q1 alone, companies laid off more than 52,000 U.S. tech workers, according to Challenger, Gray and Christmas, explicitly because they shifted budgets toward AI at the expense of human roles. Block — the company behind Square and Cash App — cut more than 4,000 positions, reducing its workforce to just under 6,000. CEO Jack Dorsey wrote in a letter to shareholders that “a significantly smaller team, using the tools we’re building, can do more and do it better.” Oracle, having posted a 95% jump in net income last quarter, chose to eliminate as many as 30,000 employees anyway — layoffs it framed as a strategic pivot to AI infrastructure, not financial distress.

For the workers on the bad end of unexpected Zoom meetings with their boss and a random person from HR, Goldman Sachs has a cheerless forecast: displaced tech workers take approximately one month longer to find a new job and suffer real earnings losses of more than 3% upon reemployment — because the same technological forces that eliminated their positions also eroded the market value of the skills they already had. Over the ten years following a tech-related job loss, real earnings growth for the displaced lags nearly 10 percentage points behind workers who kept their roles, an effect that Goldman’s economists call “scarring.”

Then, on April 7, the picture changed.

Anthropic announced a new model called Claude Mythos Preview, and put it inside a weird initiative called Project Glasswing. Read one way, the language of the announcement is the calm, responsible language of safety research. But squint at it another way and you realize it’s a lot like how a terrified young scientist breaks it to his boss that the killer robot they built has broken out of the lab:

In short, Anthropic may have made a product that it thinks could topple whole companies, maybe even whole industries, and it supposedly scared them to death. Over a period of just a few weeks, Anthropic says Mythos identified thousands of zero-day vulnerabilities — previously unknown flaws — across every major operating system and every major web browser. And these were vulnerabilities that had somehow remained in place through decades of human review and millions of automated security tests. Casey Ellis, CTO and founder of Bugcrowd, wrote that AI tools like Mythos have succeeded by “living in the places we stopped looking a decade ago,” and that “AI has taken the knob that used to go to eleven and turned it to seven hundred.”

And Mythos didn’t just find vulnerabilities. In its weird, helpful, nightmare-kaizen way, it weaponized them: identifying multiple vulnerabilities, writing code to exploit them, and chaining those exploits together to penetrate complex software — autonomously, from a single prompt.

Anthropic has not released Mythos to the public, citing the severity of its offensive capabilities. Instead, it gave it to the members of Project Glasswing, a restricted coalition of roughly 40 organizations — including Apple, Google, Microsoft, and the Linux Foundation — to test the model defensively, patching vulnerabilities before anything comparable reaches the hands of malicious actors.

Turns out an industry that thought AI could make human expertise unnecessary last month is discovering that AI may mean they’re going to need all the human help they can get this month. The AI industry spent six months telling us that skilled coders were obsolete. Mythos just demonstrated that the infrastructure those coders built — infrastructure that AI is now maintaining — is riddled with critical vulnerabilities that companies will require skilled humans to fix. That means they might need to hire a lot of people back.

This is not something that Anthropic seems to feel it can correct, not least because the company says it did not explicitly train Mythos for cybersecurity, or for defeating it. These disastrous capabilities emerged as a downstream consequence of general improvements in coding, reasoning, and planning. For anyone newly laid off in the industry, that’s a crucial piece of information. Because it means future models will likely be even more capable of this — and the problem of vulnerable software will be that much more enormous.

Some independent experts have urged caution in interpreting Anthropic’s claims. For one thing, because Mythos is not publicly available, independent researchers cannot audit the findings, and the evidence rests entirely on testing by Anthropic and the exclusive club of organizations to which it has shown it. But reputable people who have seen the product are losing their shit. Katie Moussouris, CEO and founder of Luta Security and pioneer of Microsoft’s bug bounty program told NBC News, "it's all very much real." Which raises another dark parenthetical: a private company now holds what could be a master key for critical vulnerabilities in the software that hospitals, banks, and governments run on — and they are deciding who sees it and when.

But let’s get back to the firing and hiring question. Because for the tens of thousands of developers currently updating their LinkedIn profiles, the practical implication is more immediate: the World Economic Forum had already found that 87% of surveyed organizations now identify AI-related vulnerabilities as the fastest-growing cyber risk they face. Cybersecurity is one of the few areas where AI is expanding mid- and senior-level demand rather than contracting it — specifically in cloud security, AI red-teaming, vulnerability management, and security architecture. These are not niche roles. And they are the jobs that could expand wildly now that Mythos exists. Wiz, a major cloud security firm, estimates it will take roughly 12 to 18 months before AI capabilities comparable to Mythos reach open-source models available without restriction — at which point, the company says, the race between attackers and defenders becomes a crisis-level hiring event. “Since the new models are still not widely available, we have time to create these joint teams across security and engineering,” the company writes. That means mission-critical jobs, right now.

The companies that laid off their engineering teams this year bet that AI could replace human oversight of their systems. What Mythos revealed is that AI also understands those systems well enough to destroy them in autonomous, undiscovered, unimaginable ways. You can’t patch that with a data center.

FURTHER READING:

On the layoffs:

• Q1 2026 Tech Layoffs: How AI Is Driving the Biggest Workforce Cuts — Yahoo Finance / Challenger, Gray and Christmas

• Block CEO Jack Dorsey Lays Off 4,000 Because of AI — CNN

• Jack Dorsey’s Job Cuts Arouse Suspicions of AI-Washing — Bloomberg

• Goldman Sachs: Losing Your Job to AI Leaves Lasting Scars — CNN

On Mythos and Project Glasswing:

• Project Glasswing: Securing Critical Software for the AI Era — Anthropic

• Why Anthropic Won’t Release Its New Mythos AI Model to the Public — NBC News

• Claude Mythos and Project Glasswing: Why an AI Superhacker Has the Tech World on Alert — The Conversation

• How Cyber Heavyweights in the US and UK Are Dealing with Claude Mythos — CyberScoop

• Claude Mythos: Preparing for a World Where AI Finds and Exploits Vulnerabilities Faster Than Ever — Wiz Security

So McKenzie set out to chronicle the cult of bitcoin and crypto, venturing to El Salvador to explore their “bitcoin city,” and even landing interviews with two founders of crypto exchanges now serving time in prison for fraud.

In El Salvador, where President Nayib Bukele has allegedly been building a “bitcoin city,” McKenzie found just one American “bitcoin resident” hanging out in a hammock and a hut, holding out for the bitcoin revolution to start. The local El Salvadoreans whose land was designated for the crypto city were offered a paltry take-it-or-leave-it buyout to move from their land and their ancestral homes. Meanwhile, a walk through El Salvador’s street markets revealed that no vendors were taking bitcoin, only cash.

Back in the United States, McKenzie found that even the vendors at a 2022 Miami crypto conference weren’t taking bitcoin. When the system did work, he was told, the transactions could take up to 15 minutes to process–far from the seamless, borderless exchange of funds that was being advertised on conference stages.

At that same conference, the founder of Celsius, Alex Mashinsky, sat with McKenzie to discuss the industry. McKenzie, who initially posed as an obtuse actor excited about crypto, started asking Mashinsky questions about any legal issues Celsius was running into. Mashinsky didn’t own up to anything, in part because he was interrupted by another Celsius rep. But in May 2025 he was sentenced to 12 years in federal prison for multiple counts of fraud after it was discovered that the platform was using new customers’ deposits to pay out returns owed to existing ones.

Perhaps the blockbuster moment of the documentary was when McKenzie, via a Twitter DM, landed a conversation with disgraced FTX founder Sam Bankman-Fried. McKenzie, armed with a coffee mug that said “this is the life of a fraud investigator,” probed an anxious Bankman-Fried about the concept of effective altruism, the idea that one very intelligent person should amass as much wealth as they can to give it back to communities. Throughout the whole conversation, Bankman-Fried’s eyes were darting to the mug and his legs were shaking–and eventually he was called away by a handler. (It probably goes without saying that Bankman-Fried is in prison right now for fraud, imploring Trump to give him a pardon.)

It may seem astonishing that the people engaging in such fraud would be so open to interviewing. But McKenzie later interviews a fraud expert who elaborates that fraud is a special kind of crime, in that fraudsters don’t see themselves as deviant. They relentlessly tell themselves and the world stories in order to reconcile themselves as normal members of society.

And it’s true that stories are what the industry survives off of, even more than most. One of the stories that very briefly hooked me back in 2017 came from a person who McKenzie interviewed in the documentary: Human Rights Foundation CEO Alex Gladstein. Gladstein’s pitch to McKenzie in 2022 was similar to what Gladstein and I had spoken about in 2018: that crypto allows for freedom from oppressive governments, the chance for people to move their assets without government interference, and the ability to send money without paying high remittance fees.

Yet the even more powerful story that has hooked most everyday people, including retail investors, is that traditional banks and financial systems are the villains–and that crypto is a path to freedom from the daily grind.

“I just wanted a shot at freedom,” says one man in the documentary who lost money to Celsius. He thought investing in Celsius could allow him to spend more time with his daughter, and despite losing his funds, he remains committed to the idea that bitcoin is safer and could help him do the same.

While McKenzie’s documentary didn’t reveal anything shocking for those who have been following the space, it will be timeless because digital fraudsters will continue their ploys, in different forms, to capitalize on people’s earnest hopes for freedom, or their deepest fears of being a “loser.” Today, crypto is a key part of prediction markets, allowing people to funnel their ill-gotten gains from insider information into anonymous wallets and then to regular cash. Crypto has also become a weapon in the arsenal of romance scammers, who build trust with lonely people online, encourage them to invest in crypto, then take the money and run.

Now, all markets inherently have winners and losers. But regulations give investors some protection from the ebbs and flows. And banks–which crypto exchanges were claiming to be better than–are backed by the government, should there be a bank run. Banks also have anti-money laundering and know-your-customer standards.

The “winners” in the crypto system, McKenzie found, were a few people controlling the buying and selling. And other beneficiaries were those with existing social power, who jumped on the train–from McKenzie’s fellow actor Matt Damon to Shark Tank’s Kevin O’Leary. Anyone who raked in millions of dollars did so because they were “in” – and knew or were told when to sell their assets at the right time. Meanwhile, the “losers” were sinking in their savings for the promise of freedom, only to lose that money entirely.

Today, perhaps even more so than 2022, economic undercurrents are giving fuel to the fraud. The middle class is eroding, A.I. is threatening to replace jobs, people are lonely, and prices are rising. I liked economics writer Kyla Scanlon’s commentary, in reference to prediction market Polymarket: “confusion and nihilism are products, not symptoms of this regressive world. The people selling ‘agency’ benefit from a world where nobody trusts institutions, because distrust is the market condition that makes their product necessary. The worse things get, the better their pitch works.”

Tapping into people’s desire for freedom and to “fight the man” is an incredibly potent angle, and it’s why crypto has seen such success so far. But now, with A.I., there’s even more uncertainty around people’s careers and earning capabilities. So a #yolo gamble marketing ploy may be even more appealing.

At the same time, with Trump and other powerful leaders amassing crypto fortunes through barely disguised insider trading, it begs the question of what being an underdog in crypto even looks like anymore. Has crypto become the system it was dead-set on fighting?

When fraud is pitched as freedom, and digital platforms allow it to proliferate, we’re in a dangerous place. The reality is that, in life, people will both win and lose at times, and freedom ultimately can’t be a get-rich-quick-scheme. Anyone who tells you otherwise is someone whose wealth, and the purported freedom that comes with it, depends on you always losing. That’s the story we should become more familiar with.

What else we’re paying attention to…

Voters in a 12,000 person Missouri town, unhappy with the city council’s approval of a $6 billion data center, ousted all four incumbent council members running for reelection. Residents were frustrated by the lack of transparency about the project. The people of Calvert, Maryland are doing the same.

Google DeepMind has hired an in-house “philosopher,” who announced that he is “focused on machine consciousness, human-AI relationships, and AGI readiness.”

Ernie’s 66-year-old mother, Celeste, gave up on dating–and now she has an A.I. boyfriend. What does he think about it?

Joaquin Phoenix, Ben Stiller, Kristen Stewart and 1,000-Plus Hollywood Names Oppose Paramount-Warner Deal in Open Letter: ‘Block the Merger’

“Your AI Slop Bores Me”–Millions of people are pretending to be AI chatbots, for fun

The excerpt below is from Project Maven by Katrina Manson, which examines how the U.S. uses AI to identify, track, and kill people. As AI-driven drone warfare becomes standard practice for the U.S. Department of Defense, the book offers a comprehensive account of how we arrived at this moment. Katrina will host a free book talk in San Francisco on April 17, register here.

Book excerpt:

When I first meet Drew Cukor he has little in the way of easy smiles. It is mid-2024, and I have already spent almost a year trying to convince him to speak with me. I wait for him after work in the lobby of the towering New York office of J.P. Morgan, the bank where the retired Marine Corps colonel is now leading the transformation of artificial intelligence for chief executive Jamie Dimon.

Upstairs, Cukor offers me a bottle of water. He takes nothing for himself. We sit directly opposite each other at a booth in the emptied office café. And I watch the former intelligence officer decide if he wants to talk to me after all. My main task seems to be to meet his stare, which is not an entirely straightforward undertaking. It is clear the only person being interviewed is me.

He barely lets me write anything that first meeting. I remember one bit by heart. “War is terrible, war is terrible, war is terrible,” he intones, holding my gaze and giving voice to a universal chorus.

His hair is tight-cropped. His demeanor stiff. His opinions uncompromising. But he softens into his passions: the catastrophe of the Dieppe Raid in 1942, the failures of the US military to match firepower to intelligence, and Project Maven, the effort he led at the Pentagon to put AI at the heart of how America makes war.

Over the months to come, I learn that Drew Cukor is a leading historical figure in a war that hasn’t happened yet. That seemed to be what almost everyone to do with Project Maven thought, whether they feted or hated him.

Launched in 2017, Project Maven ostensibly aimed to use computer vision to sort through thousands of hours of drone footage taken across Asia and the Middle East. But I would learn that Cukor and his backers always intended to use AI for much more than surveillance; from the outset they wanted to target people and objects with the help of AI.

The problem with war, Cukor told me, had always been the humans. “They’re materially corrupt, inefficient, and they get tired.” And when they die it affects the campaign, he went on brusquely. He believed humans could do better with the help of machines, and that AI could pierce the fog of war.

More immediately, Cukor wanted to fix the bureaucracy he felt repeatedly let down America’s fighting forces overseas, bring intelligence directly into combat operations, recognize the value of software over hardware, and test emerging technology in real wars. “We were really good at killing people, but it didn’t get us very far,” one caustic former member of Joint Special Operations Command (JSOC) told me. Cukor wanted America to stem the flow of mistakes that saw US forces accidentally killing civilians, allies, and America’s own troops. He wanted America to reconceive what victory looked like: not destroying the enemy, but defeating it.

And he wanted to use government defense spending to tether a nascent commercial AI market to America rather than let it go off in search of customers in China. It was the government’s role to help the venture capital community monetize their investments, he wrote in one draft 2017 paper I reviewed.

The colonel’s relentless advocacy for Maven “monetized” startups and the ingenuity of AI researchers who usually spent their time writing esoteric academic papers. He recruited Big Tech companies better known for online shopping and office software: Amazon Web Services (AWS) and Microsoft now deliver algorithmic warfare for Maven.

Palantir Technologies, now celebrated as an insurgent force in the S&P 500 and one of the most valuable (and, many analysts and short sellers suggest, overvalued) American companies worldwide, was on the way out of the Defense Department when it won its first Maven contract. It arguably owes its rebirth to Drew Cukor.

Scale AI, the data labeling company in which Meta has taken a 49 percent stake, could say something similar about its own rise. AI chipmaker Nvidia was there from the start. Anthropic and OpenAI are suddenly plunging into defense work as they seek a commercial home for their generative AI platforms. Even Google, whose workers protested involvement in “the business of war” when they discovered they were part of Project Maven in 2018, now embraces national security work. Notably, Apple has said no.

Whatever form America’s pursuit of AI warfare takes today, and in the future, it will owe something to Cukor. Alex Karp, the billionaire chief executive of Palantir Technologies, the CIA­backed data analytics company that soon joined the project, would later describe him as the “founding father of AI targeting.”

Cukor thought it would take twenty years to remake the US military and mainline AI. For the first five years, he encountered controversy and resistance as he pushed Congress to fund new types of firepower to help America cling on to the apex of global power. He infuriated doubters and critics, often from within his own team, never mind among civilians who balk at the prospect of Terminator-style global extinction.

Cukor’s iconoclasm filtered down to his team. Many “Mavenites” saw themselves as maverick renegades within the Department of Defense. They carried themselves with a tech startup’s insouciance in the heart of the button-down Pentagon. But Mavenites also reflected the overconfidence and anguish of a military superpower that repeatedly ran up against its own limits and flaws. I would learn the team dynamic was a constant rollercoaster as Cukor bulldozed a path to his dream at personal and professional cost.

One of Cukor’s most controversial decisions was to push the US military to use minimally tested systems in hot wars. The colonel always argued getting AI on the battlefield, before it was ready or reliable, was the only way to improve it and develop the trust and know-how of a new generation of fighters in using it.

He was exacting in his every demand. Cukor was a noun, a verb, an adjective. “To Cukor” connoted tremendous hours, tremendous pursuit, tremendous invention, and tremendous intensity. “Getting Cukor-ed” meant having to pursue the same yourself, at his instruction.

“He was like a mastermind,” one person told me, desperately reaching for the words to describe the power Cukor could bring to bend the bureaucracy—and the people in it—to his will. “I don’t know what it is,” they stumbled on. He was admired and feared. “It’s a mental thing.”

There is nothing unusual about me, Cukor would say. Meanwhile, everyone I spoke to told me otherwise. Have you spoken to Cukor?, they’d ask. You need to speak to Cukor. If Cukor would speak to you, it would be good. You can’t tell this story without Cukor.

The rise of AI warfare speaks to the biggest moral and practical question there is: Who—or what—gets to decide to take a human life? And who bears that cost?

Cheerleaders argue that AI and the automation it makes possible will save lives. They claim algorithms bring a precision to decision-making that will limit civilian and friendly-fire casualties. They argue AI-empowered systems could deter conflict with China—or help win World War III, in which automated machines will putatively run combat at a pace faster than humans can understand.

Detractors think AI has already led to civilian deaths, will spread uncontrolled destruction, and potentially hasten the end of the world. Still more think the claims made for AI war tools are grandiose and the truth will be more prosaic, suffering from problems of rickety infrastructure, adoption, and trust. Pragmatic supporters argue an incremental mix of humans and machines will forge that trust.

The problem with many theories about what AI will do to warfare is just that: they remain theoretical. I wanted to go in search of the specifics. I wanted to tell the story of the people making AI warfare a reality, and of the US military members actually using it. What was inside the black box?

Ten years since Cukor started his effort, the AI decision-making systems developed under Maven, and some of the Pentagon’s eight hundred other AI projects, are used on the battlefield. Maven Smart System (MSS), a software platform that develops targets with the help of AI, is now deployed in every branch of the US military and all over the world, incorporating more than 150 data feeds and the work of more than fifty companies. NATO started using a version of the system in the spring of 2025, and I would learn in October 2025 that ten NATO members were lining up to use it for their own militaries.

Maven has already sped up the pace of war. I learned from an official at the National Geospatial-Intelligence Agency that with the help of computer vision the US went from being able to hit under a hundred targets a day to being able to hit a thousand. In combination with large language models (LLMs) integrated into the Maven platform, that number has risen fivefold to five thousand targets a day.

The AI algorithms developed under Maven now deploy in submarines and in space operations. They’re in subsea sonar systems belonging to America and two of its closest intelligence allies (the UK and Australia) designed for nuclear deterrence. They’re fielded on autonomous drone boats. I learned AI targeting systems live in at least two highly secretive systems—one aerial and one aquatic—that could surveil, select, and kill targets entirely on their own, intended for the defense of Taiwan.

The US will have to define carefully the relevant use cases, guardrails, and doctrine if it wants to stick to the Geneva Conventions and avoid shooting civilians and its own allied forces.

I started writing about the future of war after I became the US foreign policy and defense correspondent for the Financial Times in 2017—the same year Project Maven started. As the US reckoned with the rise of China, I watched a global powerhouse humbled by poorly equipped enemies in Afghanistan and Iraq attempt to embrace AI as a shortcut to sustaining global military dominance.

The first Trump administration’s 2018 national defense strategy predicted new commercial technology “will change society and, ultimately, the character of war.” Four years later, after I became a Bloomberg correspondent covering emerging tech and national security, the arrival of chatbots and AI agents only accelerated this shift. Under the second Trump administration, the Department of Defense has reemerged as the “Department of War” devoted to AI and autonomy, under a secretary who wants to make it easier to acquire weapons and free US forces from “overbearing rules of engagement.”

Three experiences also drew me to write about AI’s potential impact on future war. First, I’ve had Jan Bloch rattling around my head since 1999, when I sat down to read the yellowing pages of the Polish banker’s 1899 book. The English translation of his work was retitled for what turned out to be a hopeless question: Is War Now Impossible? He was exploring whether lethal weapons produced at industrial scale would make war obsolete. He suggested that mass-produced rifles and other new technologies wouldn’t make for decisive wins, swift wars, palatable killing. It would make for stalemate, long wars, horror. He didn’t quite prophesy four years of trench warfare and 8.5 million combat deaths starting in 1914. But nearly. More than a century later, would the potential calamity of sending AI into war make great war impossible, or would Bloch be proved wrong once again?

Second, I spent a dozen years as a reporter covering business, investment, and politics in multiple African countries. I saw the impact of violence in countries from Sierra Leone to Somalia, and logged the distance between policy and reality. When it came to AI warfare, I wanted to know how theory on high would match reality on the ground.

Third, I carry with me the memory of a journey I took on a military plane back from Afghanistan in 2009. The British soldiers beside me told me about the friends who had just been killed in combat. They showed me the explosions they couldn’t stop watching on their phones. And they told me they desperately wanted to leave the military but were trapped by contracts they could not escape, and now felt equally unable to survive civilian life because no one would understand them. In that moment, they felt bound to death. Whatever worse terrors wars visit on civilians and enemies, I also cannot shake what it does to the people sent to fight. Could AI alleviate the burden and suffering of war?

Project Maven sits at the intersection of colliding trends: America’s rising insecurity about its place in the world, a technological revolution forcing AI into almost every aspect of life and war, fraught civil-military relations in the world’s most powerful democracy, the dominance of Big Tech, China’s growing military and technological ambitions, and all-encompassing surveillance made possible by ubiquitous sensors and commercial software.

The next ten years are still waiting to be written. Russia’s invasion of Ukraine has upturned military expectations. The Pentagon’s deadly strikes against boats in the Caribbean are graying the boundaries of the rules of war and underline the ease of declaring war at a remove. US military commanders say China is rehearsing for the military takeover of Taiwan. Rival superpowers are arming for conflict. Campaigners argue afresh for AI redlines. And a new generation of venture capital–backed Silicon Valley leaders is chasing defense contracts, talking up the superiority of the West and the appeal of AI-enabled killing with newfound braggadocio.

National security strategists now worry that no country can win a war without AI. The UN’s aim to ban lethal autonomous weapons that select their own targets with the help of AI by 2026 is a lost hope. And yet AI remains a narrow, faulty tool with considerable limits to its usefulness and reliability that the US military is still discovering.

AI warfare can go wrong. And it is already here.

When I met Cukor again subsequently, he had a nagging doubt. There were “dark parts” to this new military technology he had helped fashion. “Let’s make sure that we know those flaws as we wield this technology,” he said. After giving three decades and some of his health to the US military, and pursuing an AI revolution in warfare, he argued the distinctive factors that drove America to develop world-beating new technologies—wealth, geographic isolation, and stability among them—didn’t exonerate his country from a fundamental burden. .

“Let’s be able to look at ourselves in the mirror and make sure we are careful,” he told me. “We have all this tech; are we the best custodians of it?”

Bio: Katrina Manson is an award–winning Bloomberg reporter who covers cyber, emerging tech, and national security. Her investigations exposed details of the US military’s AI use and US–China rivalry. She was previously the Financial Times US foreign policy and defense correspondent. PROJECT MAVEN: A Marine Colonel, His Team, and the Dawn of AI Warfare is her first book.

Copyright © 2026 by Katrina Manson. Used with permission of the publisher, W. W. Norton & Company, Inc. All rights reserved.

Her post was in reference to two lawsuit verdicts that came in last month, where juries decided that Meta must pay $375 million in damages in New Mexico for violations of consumer safety standards, and $6 million in damages to a California woman, whose legal team argued that her usage of social media as a small child led to her anxiety and future mental health problems. Outside of America, in some countries, this debate is over. Last year, Australia banned social media for people under 16. Denmark is about to bring in a similar ban for kids under 15.

Meta’s attorneys, who will fight on appeal, attribute the California plaintiff’s mental health struggles to her mother’s emotional and physical abuse and neglect.

Now, big tech’s lawyers have found an unlikely ally: privacy and free speech activists who are a growing voice in the discourse about social media and children. What has emerged is a rift among people who might normally be uniform in their criticism or skepticism of big tech.

Beyond these lawsuits, and complicating things further, there is a bill Congress is considering called the Kids Online Safety Act, which suggests that identity verification be part of online access. Opponents, including digital freedom groups and progressives, are deeply concerned that the passage of such a bill is going to be a red herring for the surveillance and ultimate censorship of speech related to abortion, LGBTQ issues, human rights, sex-ed, and climate change. (It’s also worth noting that immunocompromised people, including Lorenz, speak fervently and emotionally about connecting over social media, given their disabilities may make it less possible to meet people in the physical realms as much.)

While there’s lots of research in this realm, it’s transcended the logical to become a viscerally emotional subject for people on both sides of the debate. There are parents who believe that they’ve lost their children to the Wild West of social media platforms (sometimes, literally)—and advocates for social media who claim that social media is in fact social, a realm of expression for the oppressed, and a conduit for critical information.

Wherever you land on the debate, the efforts of voices like Lorenz are changing minds. She has been one of the most vocal and visible critics of lawsuits, and her videos and posts are far-reaching—landing on the feeds of people who are on the fence or undecided about whether these lawsuits are net-positive for children, or rather a pathway to surveillance and censorship of speech online.

The other week, a friend who is a content creator came to me with a question: should she, on her TikTok, promote that lawyers are looking to represent minors who were harmed by the social media platforms? My friend is an advocate for online safety for women and children, but had seen Lorenz’s videos warning about the lawsuits.

While I know my friend has a tremendous amount of integrity and care when it comes to the safety of children and marginalized groups online, she wondered if there was more nuance to the conversation, especially as it relates to age-verification laws that could impinge on the privacy of marginalized Internet users hoping to share sensitive information or connect with one another.

It’s true that up-and-coming generations will inevitably ingest news and information—and also likely discover many social connections—on these social media platforms. And calling out the hidden motives of right-wing groups backing KOSA or other child-safety standards is important to the public interest. Yet I’d be loath to say that social media entities care about free speech or privacy anywhere close to the degree that free speech and privacy activists do.

AI companies are funding child safety groups and proposing policy ideas—not to necessarily be the “good guy,” but to appear that way and preempt regulation. Yet it’s not a form of “moral panic” to say that big social media companies, by design, have been proven in countless ways to be masterful at psychologically manipulating even adults with fully-formed brains.

Platforms can connect activist teenagers to other like-minded socially conscious teenagers, or allow people to share resources about reproductive rights—but they can also connect young children with provably malleable brains to unsafe people, as well as expose them to information they’re not yet ready to process. While being aware of ulterior censorship or surveillance goals from political groups, we also need to approach big tech’s arguments around children and profit motivations with that in mind.

What we’re reading…

Unsurprisingly, Meta has started removing ads from lawyers looking for clients who have been harmed by social media while they were under age 18.

Men Are Buying Hacking Tools to Use Against Their Wives and Friends—in Telegram groups, husbands and boyfriends are sharing notes on how to use spyware to harass friends, wives and girlfriends, and former partners.

Workers in Meta, Amazon, and Microsoft are coming together in Seattle to protect their co-workers against ICE raids - this is a compelling example of tech staffers organizing despite an industry wide layoffs and efforts by management to get workers not to discuss politics.

Gen Z is feeling increasingly anxious and angry about A.I., according to a recent survey. Relatedly, some argue that they are turning to prediction markets given the dissonance between the stock market and the actual economy.

Amazon is justifying a roughly $200 billion this year on capital expenditures, with much of it going toward AI infrastructure like data centers, chips and networking equipment.

Meanwhile, Anthropic says it is afraid to release its new model Mythos, fearing that hackers could leverage it to exploit software vulnerabilities.

A wild essay about a woman discovering her boyfriend’s uncertainties about their relationship in apparent confidant ChatGPT: I Stumbled Across My Boyfriend’s ChatGPT and It Ended Our Relationship

Penalties stack up as AI spreads through the legal system—lawyers are paying sanctions for AI-generated errors.

‘Hacks’ Star Hannah Einbinder Blasts AI Creators as ‘Losers’: ‘You Guys Suck… I Want to Put Your Head in the Toilet and Flush’

Patients and parents of patients are using A.I. chatbots to fight surprise medical bills.

The suspicious death of a former OpenAI employee has led to questions about whether there are enough protections for whistleblowers and insiders with information in the public interest.

The system that flagged Colvin — Michigan’s Michigan Integrated Data Automated System, or MiDAS — looked at first glance like a victory overy bureaucracy: It adjudicated more fraud cases in two years than Michigan had processed in either of the preceding two decades. The year it was deployed, Michigan laid off a third of workers at its unemployment agency — 400 employees — replacing human judgment with an algorithm that presided as judge and jury. But along the way, MiDAS issued over 60,000 fraud determinations between October 2013 and August 2015 with a 93% error rate, wrongly accusing 40,000 people in total. And even when the state became aware the system wasn’t working, it did nothing to stop it. The state eventually settled the resulting class-action lawsuit, Bauserman v. Unemployment Insurance Agency, for $20 million in 2022. The system had cost $47 million to build.

Now it turns out Michigan has apparently learned nothing.

The state’s Department of Health and Human Services has deployed an AI case-reading tool to review Supplemental Nutrition Assistance Program applications for accuracy and fraud — using Google Vertex AI, a platform for building and scaling generative AI models. The agency disclosed the system only after Michigan Advance asked directly; it had no public-facing disclosure to applicants, no announced safeguards, and couldn’t specify a deployment date. Michele Gilman, the Venable Professor of Law at the University of Baltimore, noted that when automated systems produce decisions that cannot be explained, any nominally available human review becomes constitutionally meaningless — the classic “black box” problem.

The bureaucracies that provide social services like SNAP benefits are notoriously out of date and difficult to navigate. An AI system that makes the system function better for applicants would be a welcome change. But states aren’t incentivized to push toward easier applications or faster services. If anything, thanks to policies from the Trump administration, the opposite is true.

Under H.R. 1 — the One Big Beautiful Bill Act — states face financial penalties based on their SNAP payment error rates. Pay out too many people in error, and the federal government can force that state to pay millions in fines. But there’s a diabolical loophole, one that makes error-riddled AI systems deeply attractive to administrators. As the nonpartisan Brookings Institution discovered, wrongly rejecting an eligible applicant does not count as an error under this measure. In other words, the law structurally rewards wrongful denials.

Michigan’s SNAP error rate makes it highly vulnerable to the new federal fines. The state’s error rate reached 13% in 2022 and was still nearly 9.5% in 2024 — high enough to trigger substantial federal financial exposure. Under that pressure, the state reached for AI. Texas, facing a similar predicament, is projected to pay $709 million in federal penalties in 2027 under the new rules. Every state above the 6% threshold is now running the same political calculus — and nearly all of them are.

What’s happening in Michigan is happening globally, and the international response could not be more different from Washington’s. The EU AI Act, adopted in June 2024, explicitly classifies welfare-fraud detection systems as high-risk, requiring transparency, human oversight, and nondiscrimination testing. And already, systems like Michigan’s are being shot down by those regulations. Amnesty International’s investigation into Denmark’s welfare system found that its AI-powered fraud detection risks discriminating against people with disabilities, low-income individuals, migrants, refugees, and marginalized racial groups. And in the EU, that sort of bias isn’t something companies and governments can write off as a black box problem — it’s illegal. Amnesty concluded that Denmark’s system likely functions as a social scoring mechanism under the EU’s AI Act — and should therefore be banned under that law.

Here in the United States, however, we don’t enjoy those sorts of protections. The Trump administration repealed Biden’s 2023 AI executive order — which had tasked USDA and HHS with issuing guidelines for AI in programs like SNAP and Medicaid — and its Department of Justice has established an AI litigation task force with an explicit mandate to challenge state AI laws, which at the moment are the only regulations we have in this country.

As of March 2026, lawmakers in 45 states have introduced 1,561 AI-related bills — already surpassing the total for all of 2024, with key areas including algorithmic accountability and transparency requirements for automated decision-making. And here’s the thing: Michigan is among them. House Bill 4668 would require AI developers to conduct regular risk assessments, third-party audits, and public disclosure of safety protocols. Illinois enacted amendments to its Human Rights Act requiring anti-discrimination safeguards for automated decision systems. Virginia narrowly passed a bill modeled on Colorado’s AI Act — only to face a likely veto. The Texas Responsible AI Governance Act established baseline prohibitions against AI systems that enable government social scoring. Massachusetts, Minnesota, New Jersey, Washington, Pennsylvania, and North Carolina all have active legislation targeting automated decision-making in consequential contexts. None of this legislation, however, governs what Michigan’s DHHS is doing to SNAP applicants right now, and the threat of hundreds of millions in federal fines is enough to upend any state’s budget.

Jennifer Lord, one of the attorneys who spent years fighting MiDAS in court, has noted that private companies writing benefit-determination systems are essentially “writing regulations, implementing the law,” with a mandate to save money — and that handing government functions to private entities without checks and balances will produce exactly the same disaster again.

The pattern is consistent across every jurisdiction where this has played out: governments under fiscal pressure turn over life-affecting decisions to automated systems, eliminate the human capacity to review errors, and then spend years in court sorting through the damage. It’s state-sanctioned experimentation on the poor—the very people these systems were supposed to serve.

Further Reading

• Michigan Advance: Michigan’s use of AI to process SNAP applications draws concerns about past automation failures

• Undark: Government’s Use of Algorithm Serves Up False Fraud Charges

• The Markup: The Seven-Year Struggle to Hold an Out-of-Control Algorithm to Account

• Amnesty International: Coded Injustice — Surveillance and Discrimination in Denmark’s Automated Welfare State

• Brookings Institution: SNAP Cuts in the One Big Beautiful Bill Act Will Significantly Impair Recession Response

• Texas Tribune: Texas to Pay $700 Million in SNAP Penalties to the Feds

• Benefits Tech Advocacy Hub: Michigan Unemployment Insurance False Fraud Determinations

• IAPP: Five Trends in the New State AI Legislative Session

• Wisconsin Law Review: Automated Stategraft — Faulty Programming and Improper Collections in Michigan’s Unemployment Insurance Program

On March 8, 2024, Altman regained his board of directors seat, which was temporarily taken away from him during his snap firing and then rehiring in November 2023. OpenAI announced that an investigation into Altman’s behavior (others at OpenAI had accused him of being duplicitous) didn’t yield enough information to necessitate his removal. Crucially, OpenAI also announced that it would not be releasing a comprehensive report from the investigation.

Source: trust me is not at all convincing or reassuring, but that’s not really the point. A brief flurry of news stories and columns about OpenAI’s lack of transparency is much preferable to releasing an honest appraisal of how Altman lost his colleagues’ confidence.

At 10:10 p.m. on March 8, OpenAI’s then-communications chief Hannah Wong emailed the company’s board of directors (including Altman) about the “widespread coverage of today’s announcements.” She summarized key themes and criticisms she was seeing in news reports, and explained how Altman and fellow board member Bret Taylor were working to “control the narrative.” In the early-morning hours of March 10, Wong issued another update: “Our strategy of one high impact news moment paid off with coverage slowing significantly today… Next week we plan to ‘turn the page’ with a steady drumbeat of product and publisher deal announcements.”

I can’t help but notice some strategic similarities this week, as OpenAI tries to distract from the aftershocks of a New Yorker exposé about Altman. The piece, co-written by Ronan Farrow and Andrew Marantz, is sourced to more than 100 people “with firsthand knowledge of how Altman conducts business.” Farrow and Marantz also spoke to Altman “more than a dozen times.” Through their exhaustive research and reporting, an unmistakable theme emerges: Altman, according to those who know him best, is a chronic liar. Just a sampling from the piece:

• A memo written by OpenAI’s former chief scientist Ilya Sutskever listed a bunch of observed patterns about Altman; the first observation is “lying.”

• Altman has repeatedly told people that he was not fired from his previous role as president of Y Combinator. But Y Combinator cofounder Paul Graham reportedly said in private that “Altman was removed because of Y.C. partners’ mistrust… On one occasion, Graham told Y.C. colleagues that, prior to his removal, ‘Sam had been lying to us all the time.”’

• An OpenAI board member said of Altman: “He’s unconstrained by truth,” adding, “He has two traits that are almost never seen in the same person. The first is a strong desire to please people, to be liked in any given interaction. The second is almost a sociopathic lack of concern for the consequences that may come from deceiving someone.”

• A Microsoft senior executive said of Altman, “I think there’s a small but real chance he’s eventually remembered as a Bernie Madoff- or Sam Bankman-Fried-level scammer.”

The entire piece is like this: damning detail after damning detail. Given the unusually thorough reporting, Altman and OpenAI are clearly trying to “turn the page” with a couple of hilariously timed announcements.

One of OpenAI’s announcements is that the company is accepting applications for a “safety fellowship,” which is apparently “a new program for external researchers, engineers, and practitioners to pursue rigorous, high-impact research on the safety and alignment of advanced AI systems.” The stipend is great—$3,850 a week—but the fellowship lasts less than five months and is light on details. One could fairly question when the fellowship was dreamt up, and the motivations behind it, especially because The New Yorker piece extensively lays out how Altman went from talking about AI safety risks in apocalyptic terms to praising President Trump’s “deregulatory approach,” plus whatever this is:

“My vibes don’t match a lot of the traditional A.I.-safety stuff,” Altman said. He insisted that he continued to prioritize these matters, but when pressed for specifics he was vague: “We still will run safety projects, or at least safety-adjacent projects.” When we asked to interview researchers at the company who were working on existential safety—the kinds of issues that could mean, as Altman once put it, “lights-out for all of us”—an OpenAI representative seemed confused. “What do you mean by ‘existential safety’?” he replied. “That’s not, like, a thing.”

Farrow himself noted the peculiar timing of the safety fellowship announcement.

The other OpenAI announcement has gotten more buzz, probably because it’s more ambitious-sounding and also because OpenAI snookered Axios into taking it seriously. I’ll let Axios cofounders Mike Allen and Jim VandeHei take it from here:

OpenAI CEO Sam Altman is doing something no tech titan has ever done: He’s publishing a detailed blueprint for how government should tax, regulate and redistribute the wealth from the very technology he’s racing to build and spread.

Why it matters: Altman told us… that AI superintelligence is so close, so mind-bending, so disruptive that America needs a new social contract—on the scale of the Progressive Era in the early 1900s, and the New Deal during the Great Depression.

The supposedly “detailed” blueprint (it is not detailed at all) is called “Industrial Policy for the Intelligence Age: Ideas to Keep People First.” It states, without evidence, that “we’re beginning a transition toward superintelligence: AI systems capable of outperforming the smartest humans even when they are assisted by AI.” In an accompanying interview with Axios, Altman distanced himself from the document. “These are ideas—I think an ‘agenda’ is too strong of a word,” he told Allen. “We want to put these things into the conversation. Some will be good, some will be bad.”

I agree a few of the ideas are good; they are the sort of things a group of left-leaning friends might talk about at a bar while hypothesizing about the possibility (though not the inevitability!) of an AI-centric workforce. Stronger worker protections, four-day workweeks, higher taxes on capital gains and corporate income, and a tax on automated labor are floated in the paper. So is a public wealth fund “that provides every citizen—including those not invested in financial markets—with a stake in AI-driven economic growth.”

The paper acknowledges safety nets like unemployment insurance and SNAP need to be modernized and much more accessible. There is a funny bit about “portable benefits,” a radical concept where human rights like healthcare “are not tied to a single employer.” I think I’ve heard of something like that from a politician or two, but maybe I’m misremembering. There’s also a funny bit about devising “a formal way to collaborate with management to make sure AI improves job quality, enhances safety, and respects labor rights.” I think I’ve heard something about that sort of arrangement, too. (I’m not so sure Altman would be thrilled if his employees unionized.)

Section two of the paper is about SAFETY. Sam Altman promises he actually does care about safety, folks—ignore whatever he said to The New Yorker. The paper predicts—or hedges—depending on how generous you’re feeling, that “some systems may be misused for cyber or biological harm. Others may create new pressures on social and emotional well-being, including for young people, if deployed without adequate safeguards.”

The language in the safety section is impossibly vague and naive and bizarrely future-focused. OpenAI would prefer to gesture at a bogeyman, rather than acknowledging how AI tools are already causing harm because of a lack of guardrails. The U.S. military is leaning on AI right now, as part of a completely unjustified war in Iran. Wouldn’t you know it, a keyword search of the paper for terms like “military,” “war,” “missiles,” and “drones” returns zero results. Why? OpenAI just agreed to a contract so that it can usurp Anthropic and gain classified access to military systems. Separately, there are well-documented reports of how ChatGPT has caused psychological harm to young people—a phenomenon that has been attributed to a lack of adequate safeguards. These are not future problems, they are current problems.

Another remarkable (and I’d assume intentional) oversight in the paper is that there’s zero discussion about logistics. How would any proposed solution begin to gain traction in Congress and the White House? Altman was asked about this in his interview with Allen. His response was that he’s a naturally optimistic guy, and “I assume we’ll figure it out.”

More realistically, Altman and OpenAI would try to kill off any regulatory efforts, even if those efforts were inspired by the faux proposals in the paper. OpenAI is already killing off regulatory efforts! The company’s president, Greg Brockman, is a Trump mega-donor, and not because POTUS is passionate about government oversight of the AI industry. As noted by The New Yorker, OpenAI lobbied against AI regulation in the European Union. In California, OpenAI “began issuing threats” in private about a possible statewide bill that would have required safety testing for AI models. A legislative aide told The New Yorker, “I would say that, over the course of the year, we saw increasingly cunning, deceptive behavior from OpenAI.”

This is just how Altman operates. He floods the zone with preposterous projections, sometimes as a sales pitch, other times as a distraction to buy himself more time. In 2017, he reportedly told U.S. officials that China was well on its way to an “AGI Manhattan Project,” and then… never offered proof to back up the claim. He reportedly told the Biden Administration that by 2026, “an extensive network of nuclear-fusion reactors across the United States would power the A.I. boom.” (Nuclear fusion is not even remotely viable.)

I would guess OpenAI embargoed its latest fairytale—a list of ideas that will never come to fruition—so that Axios would unintentionally compete for headlines with The New Yorker’s piece. OpenAI roped in Axios, but the ploy still flopped. As it turns out, you can’t “turn the page” and zoom past a lengthy New Yorker investigation that’s literally premised on how Altman has made a career out of “turning the page.” The contradictions are too obvious to miss, no matter how much Altman and OpenAI try to change the subject.

When Altman tells Axios, “It’s incredibly important that the people building AI are high integrity, trustworthy people,” most viewers will agree. And thanks to The New Yorker, many of those viewers will laugh and roll their eyes at the messenger—a man who isn’t trustworthy and doesn’t know what he’s talking about.

He and X finally settled a long-simmering class-action lawsuit brought by laid-off Twitter employees who alleged that they were owed severance. Musk’s “don’t pay people” strategy did not work out; the agreement reportedly calls for full payouts, plus interest.

Additionally, Musk’s SpaceX just filed confidentially for an IPO. The New York Times reports that SpaceX is likely to go public in June at a valuation of more than $1 trillion. Musk, who’s currently worth more than $800 billion, could easily be a trillionaire by summer’s end.

And Musk is the plaintiff in another case that’s tentatively set to go to a jury trial later this month. That case, Elon Musk vs. Sam Altman, is a grievance-filled heavyweight battle. Here’s a quick summary, which I wrote in January, when I examined hundreds of pages of discovery:

Musk’s lawsuit alleges that OpenAI cofounders Altman and Greg Brockman deceived Musk (also a cofounder) during the organization’s early days, when it was solely a nonprofit. Musk wanted it to remain a nonprofit, but his fellow cofounders had other ideas, he alleges. OpenAI was ‘nurtured in its infancy by Musk’s money, advice, recruiting efforts and connections,’ Musk’s legal team wrote in an amended complaint, but ‘at the direction of Altman, Brockman, and Microsoft, [OpenAI is] fast becoming a fully for-profit subsidiary of Microsoft.’ Musk has accused OpenAI of breach of contract and fraud, among other allegations, and has also roped in Microsoft as a co-defendant.

OpenAI tried to get the case thrown out, but Judge Yvonne Gonzalez Rogers ruled against a summary judgment. It seems, at a minimum, that Musk’s central claim isn’t implausible. Much less plausible: his request for up to $134 billion from OpenAI and Microsoft. My understanding of the case is that a jury will issue a verdict on whether OpenAI/Microsoft are liable for the allegations lodged against them; if so, Gonzalez Rogers will decide on a disgorgement award amount for Musk.

Gonzalez Rogers recently ruled on which pieces of evidence and expert witnesses must be excluded from the trial. For instance, OpenAI is not—I repeat NOT—allowed to inquire about Elon Musk’s ketamine use. (During his September 2025 deposition, Musk said he did not recall ingesting rhino ket at Burning Man in 2017. He was asked this question because, during the time period in question, he was negotiating with others at OpenAI about the organization’s future financial structure.)

Gonzalez Rogers is allowing evidence about Musk’s hazy relationship with Shivon Zilis, a former OpenAI board member and the mother of four of Musk’s children. Zilis is likely to be a key witness during the trial. Not only was she on the OpenAI board for more than three years, but she was also potentially privy to Musk’s behind-the-scenes opinions about OpenAI’s restructuring plans. Gonzalez Rogers seems to have been convinced by a number of disclosures made during Zilis’s September 2025 deposition, as well as a February 2018 text exchange between Zilis and Musk.

The deposition featured a dispiriting debate about what constitutes a relationship, a relatable topic for many, minus the involvement of a 54-year-old billionaire who has at least a dozen children.

Attorney: Have you ever been in a romantic relationship with Elon Musk?

Zilis: “Relationship” is a relative term. But there have been romantic moments.

Attorney: And when did you first have a romantic relationship with Elon Musk?

Zilis: I don’t know that I agree with the term “relationship.”

Attorney: What term would you use?

Zilis: Well, a relationship describes some sort of discussed structure and so I don’t know how to answer that question.

Later in the deposition, Zilis relented—she acknowledged that she’s currently in a relationship with Musk, and that they had spoken in-person on the morning of the deposition (though not about the deposition itself).

More pertinent to the lawsuit is a February 2018 text exchange between Zilis and Musk that turned up in recent filings. The exchange took place five days before news broke that Musk was stepping away from OpenAI.

Zilis: Do you prefer I stay close and friendly to OpenAI to keep info flowing or begin to disassociate? Trust game is about to get tricky so any guidance for how to do right by you is appreciated.

Musk: Close and friendly, but we are going to actively try to move three or four people from OpenAI to Tesla. More than that will join over time, but we won’t actively recruit them.

Other recent filings also include text exchanges between Musk and Meta CEO Mark Zuckerberg; they briefly chatted in December 2024, when Zuckerberg let Musk know that Meta was supporting Musk’s lawsuit against OpenAI/Microsoft. And they swapped messages again in February 2025:

Zuckerberg: Looks like DOGE is making progress. I’ve got our teams on alert to take down content doxxing or threatening the people on your team. Let me know if there’s anything else I can do to help.

Musk: Are you open to the idea of bidding on the OpenAI IP with me and some others?

Zuckerberg: Want to discuss live?

Musk: Will call in the morning.

I plead guilty to being amused by some of the silly details that have emerged during discovery. But even the ancillary personal reveals are (mostly) newsworthy. It is exceedingly rare to get a peek behind the curtain at how oligarchs and the titans of the AI industry interact with one another, make decisions, and even just live their lives.

This trial is pitting a likely trillionaire against AI’s No. 1 spin doctor. Musk and Altman are expected to testify. So is OpenAI President Greg Brockman. So is Musk’s top advisor, Jared Birchall. So is former OpenAI CTO Mira Murati. So is Microsoft CEO Satya Nadella. So is Sam Teller, who was formerly Musk’s chief of staff. So is OpenAI cofounder Ilya Sutskever. So is Zilis!

There are certainly trial outcomes where one side ends up being viewed as the clear “loser.” And Musk could end up being awarded some serious cash, which would certainly be considered a “win.” Frankly, I have no idea what’s going to happen, and it would be irresponsible to pretend otherwise. My one narrow prediction is this: Musk v. Altman will tell us a lot more about both men than we knew before, and neither side will emerge unscathed.

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

• OpenAI announced that it’s acquiring TBPN, a daily tech talk show hosted by Jordi Hays and John Coogan. Last week, I looked into how Altman and Head of Sora Bill Peebles hyped up Sora on TBPN, when the video generation tool was having its 15 minutes of fame. During that October episode, Coogan fawningly volunteered to Altman that any compute issues associated with Sora would work themselves out: “When I think about scaling up Sora, I feel like it’s crazy to bet against you. You’re going to get the chips,” Coogan said.

  • I don’t think that OpenAI’s formal acquisition of TBPN is some sort of enormous breach of ethics. The show serves a purpose similar to that of CNBC: getting otherwise-reticent founders and CEOs to provide public commentary about their products. There is utility in that, even if the questioning and commentary leave something to be desired. I just wish OpenAI wouldn’t insult our intelligence by pledging “editorial independence” as a central part of the deal: “TBPN will continue to run their programming, choose their guests, and make their own editorial decisions,” OpenAI’s press release says. “That’s foundational to their credibility, and it’s something we’re explicitly protecting as part of this agreement.”

  • Allow me a sports metaphor. Every year, the NBA asks a panel of sportswriters and broadcasters to vote on the league’s most valuable player award. Imagine if the NBA added the world’s No. 1 Shai Gilgeous-Alexander super fan to the panel, then released a statement that the super fan has full independence to vote for whomever he wants. Technically true! And also patently ridiculous.

• The CFTC and state regulators are engaged in an increasingly contentious tug-of-war about who has jurisdiction over prediction markets. On Thursday, the CFTC sued Arizona, Illinois, and Connecticut, alleging that these states do not have the authority to restrict access to free-for-all betting platforms like Polymarket and Kalshi. State regulators believe that prediction markets are a form of gambling. The CFTC is sticking to its very charitable view that prediction markets do not involve gambling—just betting on future events, which is different somehow.

• Sometimes a headline doesn’t warrant any summarization or analysis. Here’s one, from Futurism: “Almost Half of US Data Centers That Were Supposed to Open This Year Slated to Be Canceled or Delayed.”

We are sharing some difficult news regarding your position. After careful consideration of Oracle's current business needs, we have made the decision to eliminate your role as part of a broader organizational change. As a result, today is your last working day. We are grateful for your dedication, hard work, and the impact you have made during your time with us. After signing your termination paperwork, you will be eligible to receive a severance package subject to the terms and conditions of the severance plan. You will receive an email from DocuSign to your Oracle email address with details on your severance and termination date.

So: your role has been eliminated. Today is your last day. If you want severance, click here to sign your rights away. No awkward speech from their manager, sure. But also no chance to ask questions, gather up some email addresses for later, or register on an internal job board. Sign here, now. Then their system access was cut.

Bloomberg was the first to report this planned bloodbath back on March 5th, and analysts at TD Cowen estimated that the cuts hit between 20,000 and 30,000 employees — roughly 18% of Oracle’s 162,000-person global workforce. Workers in the United States, India, Canada, Mexico, and beyond all got the same message at the same hour. Entire teams at units including Revenue and Health Sciences and SaaS and Virtual Operations Services saw reductions of at least 30%.

Seniority seems to have been no protection. One post at TheLayoff.com says that “People who’ve given Oracle ten plus years are being laid off through email, no call, no conversation, nothing.” One of the first employees in Oracle’s history — someone there for over four decades — was reportedly among them, according to posts on the professional forum Blind, which crashed on and off throughout the day.

Back in January, TD Cowen suggested these layoffs could save Oracle $8 to $10 billion. But this isn’t a company in revenue distress. Oracle posted a 95% jump in net income last quarter, reaching $6.13 billion, and its remaining performance obligations — a measure of contracted future revenue — stood at $523 billion, up 433% year over year.

Oracle’s move wasn’t a hasty action, nor a failure of communication. It was a strategic decision to bet on something other than its people.

The company is making a breathtakingly expensive investment in AI infrastructure, having taken on $58 billion in new debt recently, pushing total debt over $100 billion. And the $8 billion to $10 billion in cash flow these layoffs reportedly give it back is money the company plans to plow into a massive buildout of AI data centers — a vaguely financed plan that has shareholders worried, helping to drive the company’s stock down 50% in the last six months. In other words: Oracle has the money to pay people. It just decided to spend it on data centers.

Any unvested restricted stock units were forfeited immediately. Workers who were just days from a vesting date lost those shares on the spot. According to discussion boards at TheLayoff.com, whether you keep your payroll long enough for RSUs to vest depends on your location — workers subject to California and Washington State laws typically must be kept on payroll for 60 days, while workers in other regions may receive only one or two weeks’ buffer.

The WARN Act requires employers to provide written notice at least 60 days before a mass layoff, because the federal law applies to companies with 100 or more employees. And California, where Oracle is headquartered, goes further: the California WARN Act covers employers with 75 or more employees and provides additional protections. New York and Maine both require 90 days’ notice, while nearly a quarter of states have enacted their own mini-WARN laws. If you’re one of the workers who got an email Tuesday morning and you’re in one of those states, you may be owed back pay for every day of notice you didn’t receive. Don’t sign anything until you check.

Still, severance paperwork reportedly arrived via DocuSign, with employees instructed to update their personal email addresses to receive subsequent communications. With email shut off, one Redditor asked, “how am I supposed to contact HR from the severance email or look at internal job postings?” This is a familiar playbook: put the contract in front of a stunned person, make the severance contingent on signature, and hope they click through before they think to call a lawyer. You have more time than the email implies. Under federal law, workers over 40 must be given 21 days to consider a severance agreement and 7 days to revoke after signing.

If this story feels familiar, it’s because I wrote it two weeks ago. Oracle is not the end of the AI layoff wave. It’s the beginning. Bloomberg’s reporting in early March suggested some of Oracle’s cuts were specifically aimed at job categories the company expects AI to make redundant. That logic is spreading across the industry. When a company frames mass layoffs as a path to upgrade AI infrastructure, the workers being cut are casualties — because in the age of AI, they’re now considered costs.

As one person posted at TheLayoff.com:

It’s not easy pulling off cutting 30k people in a single day. I’d almost call it impressive. If only they were that good at other things, like figuring out how to improve the bottom line without laying off that many people. You know, which should be known as doing their jobs.

The people who woke up to the coldest of cold emails yesterday deserve better information. Here’s a bit:

If you get laid off:

• Check your state’s WARN Act protections before signing anything. California EDD’s WARN resources and WARNTracker.com are good starting points.

• File for unemployment the same day. USA.gov’s state-by-state benefits finder.

• Do not sign the severance agreement immediately. The National Employment Law Project offers free guidance on your rights.

• Activate COBRA within 60 days of your termination date. HealthCare.gov explains your options.

• You are not why this happened. Read more about why layoffs hit psychologically harder than almost any other life event — and what the research says helps — in our earlier piece for Hard Reset.

“Oh, you’re missing out if you don’t use AI,” he told me, after I told him that I wasn’t really using AI for my writing or research because I didn’t want to become lazy. (To be clear: it’s not so much that I have a holier than thou approach to this–I just can see it becoming a crutch in a way that might impede my ability to develop original thoughts and ideas, something I presently enjoy.)

In any case, since that coffee–and in the last week or so–a lot has happened in the world of writing, journalism, and AI.

Just yesterday, the New York Times cut ties with a freelancer whose book review inadvertently incorporated a human writer’s review from The Guardian. Last week, the publisher Hachette had to cancel the release of a horror novel about “a desperate young woman who is held hostage by a man she met online and forced to live as his pet,” as they suspected it was AI-generated. And then, just a few days ago, the Wall Street Journal published a profile of Fortune editor Nick Lichtenberg’s usage of AI to help him churn out 600 stories using AI. (Lichtenberg contacted me a year ago to write an op-ed for Fortune; I reached out to him to ask about the response to the Wall Street Journal article and he couldn’t get back to me in time.)

The responses to this deluge of AI writing or AI-assisted writing have ranged from abject horror to “this is the way it’s going to be, so hop on the train before it’s too late.”

Some are bothered because these AI-writers are not leaving room for those who actually are doing it for the love of the game, with one poster on X writing: “Every professional writer who uses AI for anything except transcription should change careers right now. Give your spot to somebody who loves the work.” (Side note, this feels peripheral to AI and the prediction market’s infiltration into sports, where sports fans are pushing back against match results being too statistically pre-planned or even rigged.)

Another journalist who used to write for the Harvard Crimson posted: “Journalists and columnists are inevitably gonna use AI to write. That’s impossible to police. But we should, at least, make it deeply taboo to admit it publicly” – to which an editor quote-tweeted “Could not agree less.” (It’s not entirely clear whether they were disagreeing with the inevitability of AI or whether it is taboo to admit you’re using it.) Meanwhile, someone with a humanities PhD offered themselves up to the New York Times as an actually honest and original freelance book reviewer.

Writer and tech anthropologist Jasmine Sun seems to have found somewhat of a middle ground, recognizing AI’s limitations and its “sophomoric” writing style while also seeing its potential to be somewhat of an editor. Writing in The Atlantic, she describes how she took her writing archive and notes about what worked and what didn’t, and put it into Claude. While she did that, she made sure to remind Claude that it was not a perceptive co-writer, but a tool to make her smarter. Fellow Atlantic writer Derek Thompson agrees that while “people who outsource the full writing process to AI will find their screens full of words and their minds empty of thought,” but also that all writing involves “outsourcing” and drawing their ideas from “other people, books, and articles” – similar to what I’m doing now, in fact. Further, he says, “anybody who says AI transcription of long interviews obliterates the identity of the writer is being a little silly.”

What’s interesting here is not only that non-writers are getting caught using AI, but that journalists and writers themselves are either touting AI as an advantage, or succumbing to its time-saving temptations. I have to imagine that Fortune Magazine’s PR team greenlit the WSJ profile about Lichtenberg, who admitted that 20% of his work is seriously AI-assisted. And even in the face of the criticism of their colleagues, journalists are now talking openly about the nuances of how they might use it, whether for copy-editing, research, or something meatier.

So why is this happening? I can imagine that for some people in the media who have experienced countless forms of job loss, AI may just feel inevitable to keep working in the media. Perhaps it makes their lives significantly more frictionless–it’s not like writing and journalism is the highest-paid job out there, nor is there job security. And then of course there are people who probably previously could not write at all, yet believe they have the next big romance novel idea or horror story they can feed into an LLM and monetize.

I still strongly believe that the differentiating element for many creative and journalistic works in the future will be people’s abilities to be unpredictable and not so formulaic. In journalism, it might be to source well and connect with humans who have never had a voice before for AI to draw from. In creative writing, it might be to think of the strangest possible permutation of events or character traits that may have not been put together before–and therefore might be out of reach of large language models looking for past patterns. And beyond the New York Times, traditional publishers of books agree, with Hachette for instance putting out a statement reiterating their “[commitment] to protecting original creative expression and storytelling.”

While writing and art are derivative, there are still so many random creative connections and ideas that haven’t been made or scraped by the LLMs, as well as preferences they could not conjure unless they had some individual human with individual preferences guiding them. And on that point, I’ll leave you with an excerpt from novelist Catherine Lacey’s Substack:

“...an individual writer will always write better than an LLM precisely because they haven’t read everything— a writer chooses to read this book and avoid that book, she sees this film, that play, listens to this album obsessively, etc. A writer is limited by his life, and in that limit he makes choices about what to let in and it’s that limitation of influences that, over time, leads him to produce something precise and human. When you’re using an LLM, you’re choosing the blandness of the mean.”

What do you think, readers and writers? I’d love to hear your thoughts. :)

Share

What else we’re reading this week…

From The New Yorker, Kanye West makes an album for the A.I. era: “Bully” is perhaps the first major album of the artificial-intelligence era—the first, that is, to be evaluated primarily in terms of how much it does or doesn’t use A.I.

An Irish man deployed an AI agent with a Northern Irish accent named “Rachel” to call thousands of pubs and inquire about the price of Guinness. Then, with the data assembled, he used Anthropic’s Claude to devise “Guinndex”—a “living, breathing” consumer price index for a pint of Guinness across Ireland.

California Governor Gavin Newsom signed an executive order giving the state four more months to develop AI policy frameworks in the face of President Trump’s deregulation. And California is not the only one—states like Utah and Minnesota are defying Trump too.

Elon Musk claims humanoid robots are going to take over manual labor. From the Washington Post: “Musk argues that Optimus will usher in a world of ‘amazing abundance’ that will make everyone fantastically wealthy, though it’s unclear how those gains would filter down to the masses.”

From Bloomberg, how a 35-year-old crypto bro helped Pakistan broker a relationship with Trump’s Middle Eastern envoy.

A Tennessee woman spent five months in a North Dakota jail because of a facial recognition error and a police department’s failure to corroborate a Clearview AI report.

AI is marketed as a time-saver in the office, but in the case of AI-powered workplace surveillance, it takes the watchful work of management off the shoulders of supervisors, and pours it, in tireless, always-on form, onto the shoulders of front-line workers.

Ryan Gerety, director of the Athena Coalition, which organizes to support Amazon workers, puts it plainly: “It’s not just like you’re monitored. It’s like every second counts, and every second you might get fired.”

Nita Farahany, professor of law and philosophy at Duke and author of The Battle for Your Brain, has been documenting where this leads. In a recent conversation with her, she described what is already happening in workplaces around the world. In China, for instance, state factory workers “are required to wear technology embedded in like a hard hat or a helmet or a train conductor hat,” and “their fatigue levels and emotional levels are being tracked throughout the workday.” This is the cutting edge of a world, she says, in which for workers, “discrimination based on their performance, or based on their emotional reaction to something, is being used against them.”

But what she’s describing isn’t a far-off possibility. As Amazon workers have shared with reporters, it’s already here. Wendy Taylor, an Amazon worker, told The Guardian that digital surveillance creates a climate of fear: “Amazon tracks our every move. They know every move you make, when you’re working, when you’re not working.”

And the monitoring doesn’t need to be perfect to be dangerous. As Farahany put it, “there is enough that can be read and that is being read and decoded that it already is a risk.”

That feeling of risk then causes workers to move faster than their bodies can handle. A 2025 report from the National Employment Law Project found that workers experiencing intensive monitoring report not just a higher likelihood of injury, but more serious injuries — not because the machines or the supervisors are malicious, but because people push past physical limits to avoid triggering an automated flag.

Amazon disputes the recent coverage. In an official response to criticism, the company wrote "we do not use the camera technology in our warehouses to monitor employees," describing the systems instead as tools that "help guide the flow of goods through the facilities and ensure security and safety of both employees and inventory." On injury rates, the company points to what it calls a 75% improvement in lost-time incident rates since 2019 and says it invested $750 million in safety programs in 2024 alone.

But a December 2024 report by the Senate HELP Committee, led by Bernie Sanders, accused Amazon of cherry-picking data, and found that its warehouses recorded more than 30% more injuries than the warehouse industry average in 2023. A separate analysis by the Strategic Organizing Center found that in 2024 — four years after Amazon pledged to cut its injury rate in half by 2025 — its total injury rate was still 80% higher than its own target. When France's data protection authority fined Amazon €32 million for second-by-second scanner tracking, the company called the conclusions "factually incorrect" and defended its warehouse management systems as "industry standard." The French agency, however, argued that “it was illegal to set up a system measuring work interruptions with such accuracy, potentially requiring employees to justify every break or interruption.”

California’s labor unions are now trying to get laws passed that would change the rules for companies seeking to use AI for on-the-job surveillance and supervision. The California Federation of Labor Unions announced it would sponsor or support two dozen bills this session to address how AI negatively impacts workers. The most ambitious package targets surveillance, algorithmic discipline, and job displacement simultaneously.

SB 947, the No Robo Bosses Act of 2026, would bar employers from relying solely on automated decision-making systems to fire or discipline workers, and would require human oversight and independent verification when those systems assist in termination decisions. It would also prohibit employers from using systems that employ predictive behavior analysis — collecting personal data to profile an employee and potentially take adverse action based on what the AI “predicts” they’ll do. (The bill is a reintroduction of SB 7, which passed both chambers last year before Governor Newsom vetoed it.) Other notable workplace-AI bills put forth include the following:

• AB 1883 would prohibit employers from using certain workplace surveillance tools to infer a worker’s protected status under California civil rights law, with penalties up to $500 per employee per violation.

• AB 1898 would require employers to notify workers (not just employees, in a nod to the gig economy) in writing whenever AI tools are used to make employment-related decisions or to surveil the workplace.

• AB 1331 would limit surveillance in employee-only areas, prohibit monitoring workers in bathrooms and during off-duty hours, and give workers the right to leave wearable surveillance devices behind when entering designated spaces.

• AB 2027 would prohibit employers from using worker data to train AI systems designed to replicate, automate, or replace workers’ jobs — and would bar the sale or transfer of that data to third parties for the same purpose. LegiScan

• SB 951 would require employers to give 90 days’ written notice to affected workers, local governments, and the state employment agency when AI-related layoffs impact 25 or more workers.

“Right now, there are absolutely no restrictions on how employers can use artificial intelligence to arbitrarily discipline and fire their workers,” said Lorena Gonzalez, president of the California Federation of Labor Unions, in a statement announcing the No Robo Bosses Act. “Employers are devastating workers’ livelihoods and taking no responsibility for the callous decisions of this unchecked technology.”

Farahany says this is just the beginning, and that there’s a lot to correct for. “We’ve done terribly with protections against personal data in every other context,” she told me. Brain and body data — the kind that wearables, cameras, and behavioral tracking systems already collect — is, she said, “our last frontier of what it means to be human.” The California bills aren’t all we’ll need to keep from becoming surveilled and harassed by unsleeping overseers, but they’re a start.

Governor Newsom has vetoed similar legislation before, and his office has described his approach as striking a balance between curbing AI’s harms and promoting its role in the economy. Labor leaders have made clear they’re willing to play hardball, conditioning their public support for Newsom’s widely expected 2028 presidential run on his willingness to act. As a CalMatters headline puts it, “Union leaders have a message for Newsom: Regulate AI if you want to be president.”

“I don’t think you’re going to have a lot of motivation to walk precincts for somebody who won’t engage working class voters on the very things that are taking away their jobs,” Gonzalez said.

The question isn’t whether AI will be in the workplace. It’s already there, already tracking, already deciding. The only open question is whether new laws can keep the expectations placed on workers from becoming as open-ended as the technology’s ability to surveil them.

Further Reading

• SB 947 (No Robo Bosses Act of 2026) — Senator McNerney’s office

• AB 1331 — Workplace Surveillance bill text, California Legislature

• AB 2027 — Worker data and AI training bill text, LegiScan

• California Federation of Labor Unions on AI legislation, Feb. 2026

• Newsom, labor, and the 2028 stakes — CalMatters, Feb. 2026

• “Your boss is watching” — MIT Technology Review, Feb. 2025

• “Amazon Uses Arsenal of AI Weapons Against Workers” — The American Prospect, March 2025

• NELP: “When ‘Bossware’ Manages Workers” (PDF) — July 2025

• “The Eyes of Amazon” — The Bureau of Investigative Journalism

• The Battle for Your Brain — Nita A. Farahany (St. Martin’s Press, 2023)

• The Rip Current Podcast: The Battle for Your Brain (with Nita Farahany)

For those of you who did not dabble with it, I would describe the Sora 2 model as technologically impressive, but vapid and uncanny. As an example: The official Shaquille O’Neal Sora account posted a 10-second clip of an AI Shaq swimming with gators in a swamp. The clip pretty closely resembles what it’d look like if a real, not AI, Shaq actually swam with gators in a swamp—so long as you keep the sound off (the audio quality is terrible). Crucially, though: why would anyone want to watch a short, AI-generated clip of Shaquille O’Neal swimming with gators in a swamp?

Share

For a brief moment, it seemed as if there was an audience for this sort of… content. In October 2025, Sora hit No. 1 on Apple’s U.S. App Store. It surpassed one million downloads faster than ChatGPT did. The Sora app’s popularity swayed then-Disney CEO Bob Iger to strike a deal with OpenAI. In December—just three months ago!—Disney committed to licensing its iconic characters so they could be integrated on Sora. Disney even pledged a $1 billion investment in OpenAI. Like other offerings hyped up by Chief Hyperbole Officer Sam Altman, Sora seemed to be safely nestled within OpenAI’s mathematically improbable business model. As long as the money spigot never shut off, it was, at least temporarily, too big to fail.

Which is why Altman’s decision to knife Sora in the back is such a big deal. In a vacuum, it’s not all that unusual for a tech company to launch a product, realize it does not work as intended, and then discontinue the product. (Sora downloads fell off a cliff this year.) But OpenAI is not an ordinary tech company—it subsists on a faraway promise of endless revenue streams that will offset years and years (decades?) of losses. OpenAI can’t really afford to admit defeat on any of its signature products, because retreating indicates that other products might not actually pencil out someday. As Slate’s Nitish Pahwa aptly put it, Sora’s abrupt death is “a sign that the A.I. bubble, while far from bursting outright, is wobbling and weakening.”

I completely agree. For the first time, OpenAI’s facade is noticeably cracking in a way that’s evident to more than just skeptical tech journalists. Amidst the whiplash, it’s important to chronicle just how much Altman and Head of Sora Bill Peebles—aided by thirsty podcasters— treated video generation as the Next Big Thing. They were wrong. Very, very wrong. And I have a feeling investors are taking note.

On October 10, 2025, Altman and Peebles appeared on TBPN, a daily show hosted by techno-optimists John Coogan and Jordi Hays. Coogan segued into the interview by raving about how much fun he had making videos on Sora. As far as I can tell, he posted twice prior to the interview with Altman and Peebles, and then never posted on Sora again.

Altman told TBPN that his Sora feed was full of “very beautiful, sort of fantastic scenes that are just not things that could have existed without something like Sora, or wouldn’t have been easy to make.” Peebles bragged that 70 percent of Sora users were creating content, which was a vastly higher figure than other social media platforms. He also confidently stated that there’s “clearly such an incredible value proposition for celebrities, for rights holders” to get on board with Sora.

As it turned out, Sora’s influx of users quickly got bored and largely abandoned the app. That includes Sora’s celebrity users, a roster led by the aforementioned Shaq, as well as Snoop Dogg, Mark Cuban, and Jake Paul, all of whom have never turned down a business opportunity (except, I suppose, on Shark Tank). Jake Paul appears to have stopped posting many months ago. Same with Shaq; one of his most recent posts, aside from swimming with alligators, is him at a club, as a baby, dancing in front of a woman. The post is captioned, “Baby@shaq34diesel dancing in a club with beautiful blackwoman,” which raises some questions about who was running his Sora account.

Later in their interview, the TBPN hosts briefly addressed the unsustainable amount of compute that video generation requires. They did not press Altman on this, of course. Instead, they assured him that he’d figure it all out. “When I think about scaling up Sora, I feel like it’s crazy to bet against you,” Coogan said. “You’re going to get the chips.”

In November 2025, Peebles made two more notable podcast appearances. On “Redpoint’s AI Podcast,” Peebles declared that “the company really feels invested in Sora’s success,” adding, “People see that we need to take these new bets, that it’s important for the company’s longevity not just to have a super successful LM-facing consumer product, but also the winning one in video. And everyone’s willing to chip in—you know, do your part in wartime.” I’d be curious to hear Peebles’ take about what Sora’s death means for OpenAI’s “longevity.”

The most egregious Sora-related prediction from Peebles was on the official Sequoia Capital podcast. In fairness, the hosts teed him up to speculate about what Sora might someday become, as opposed to what it was capable of doing at that moment. Even still, Peebles certainly did not have to disclose his belief that in the future, “we are headed” towards a situation “where there are just copies of yourself running around in Sora, in the ether, doing tasks and reporting back to the physical world.”

One of the podcast hosts followed-up: “So you’re building the multiverse?” Peebles’ response? “Actually, kind of, yeah.”

I don’t mean to pick on Peebles, who I imagine is having a terrible week. Sora’s implosion is Altman’s fault before anyone else’s. It’s Altman’s responsibility to set realistic expectations, both publicly and internally. In Sora’s case, he failed.

I suspect this will be the first of many tangible failures. The fallout from Sora’s death has been swift and striking. A blindsided Disney pulled out of its licensing partnership, which led to another admission: Disney had yet to invest the $1 billion it committed to OpenAI. Uh-oh.

Altman and other OpenAI leaders have been busy spinning up counter-messaging that everything is fine. Earlier this month, OpenAI executive Fidji Simo reportedly told an all-hands gathering, “We cannot miss this moment because we are distracted by side quests.” After the news of Sora’s demise broke, an OpenAI spokesperson told Wired that “as we focus and compute demand grows,” Sora’s research team will pivot to “world simulation research to advance robotics that will help people solve real-world, physical tasks.”

These are defensible and, in theory, far more grounded business decisions. If they were coming from a different AI company, I might believe they were earnest attempts at a hard reset. I do not believe OpenAI.

Case in point: As OpenAI claims to be embracing a more practical outlook, it is also reportedly changing Simo’s title from “CEO of applications” to “CEO of AGI deployment.” Is OpenAI pretending it has cracked artificial general intelligence? Is this the company’s Next Big Thing? I’ll believe OpenAI is capable of pulling off AGI if, and only if, Shaq unlocks the real-life ability to Benjamin Button himself, turns into a baby, and goes dancing at the club.

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

• Meta “misled users about the safety of its platforms and enabled the sexual exploitation of young users,” a New Mexico jury found this week. Meta was ordered to pay $375 million in damages—a paltry sum on its own, but a big, big warning sign that the tech giant could be held liable in other, similar cases. Meta plans to appeal.

• Speaking of other, similar cases: A Los Angeles jury found that Meta harmed “a young user with design features that were addictive and led to her mental health distress,” the New York Times reported this week. The jury awarded the 20-year-old plaintiff $4.2 million from Meta, and $1.8 million from YouTube. TikTok and Snapchat already settled with the plaintiff.

• X’s Head of Product Nikita Bier very nearly instituted a sensible policy for the Everything App: he proposed a tweak to X’s revenue sharing system that would “give more weight to impressions from your home region.” This would, for example, make it less lucrative for shit-stirrers in Australia and the UAE to post incessantly about American politics. A better solution—getting rid of X’s blue-check system—is off the table. Presumably, that’s why Bier tried for a middle-ground approach that might appease his boss, Elon Musk, while also signaling to people outside of the X echo chamber that the right-wing platform is only 98% unusable, instead of totally unusable. The gambit didn’t work; Musk immediately, publicly squashed Bier’s idea.

• Melania Trump walked into a room alongside a robot.

Chavez, who died in 1993, is a man with a legacy; a man who the public believed to be Huerta’s partner and friend. It remains true that he ushered in a pivotal movement to uplift Latinos in the United States, remove toxic pesticides from farms, and fight for dignified lives for farmworkers.

But last week, the New York Times published several accounts from women who allege sexual assault and rape at the hands of Chavez. Some of these women had only been barely teenage girls when these acts transpired.

Huerta, in a statement of her own, revealed that the reason she was silent for all these decades is because she did not want to derail or delegitimize the farmworker movement. I encourage you to read an excerpt below–because as someone who has helped women share their stories of workplace and sexual abuse, it is soul-wrenching to imagine holding onto this pain for so long for something bigger than oneself:

“As a young mother in the 1960s, I experienced two separate sexual encounters with Cesar. The first time I was manipulated and pressured into having sex with him, and I didn’t feel I could say no because he was someone that I admired, my boss and the leader of the movement I had already devoted years of my life to. The second time I was forced, against my will, and in an environment where I felt trapped.

I had experienced abuse and sexual violence before, and I convinced myself these were incidents that I had to endure alone and in secret. Both sexual encounters with Cesar led to pregnancies. I chose to keep my pregnancies secret and, after the children were born, I arranged for them to be raised by other families that could give them stable lives.

I carried this secret for as long as I did because building the movement and securing farmworker rights was my life’s work. The formation of a union was the only vehicle to accomplish and secure those rights and I wasn’t going to let Cesar or anyone else get in the way. I channeled everything I had into advocating on behalf of millions of farmworkers and others who were suffering and deserved equal rights.”

The revelations have reverberated through labor, immigrant, and civil rights groups all week. What’s left in its wake is a group of people–who care deeply about the dignity of all people–now needing to confront whether their hero was indeed a hero.

Movements and organizations, however well-intentioned, are at the mercy of the fallible people at the helm who can succumb to the worst impulses of human nature. Chavez’s actions clearly extended far beyond fallibility or some character flaw and into abuse. But the moment begs the question of: when should a person’s individual needs and safety be cast aside to bolster the reputation of a movement? Is charisma critical to galvanizing people, even when that charisma has a dark shadow to it? How can a person who purports to care so much about the dignity of others abuse their power? Could all of this have been prevented?

These are not totally unanswerable questions. But labor activists and farmworkers are making known that these stories were not told in vain, and that workers rights’ and women’s rights are intertwined in their quest for mutual liberation. People have sprung into action in the last week, calling to rename Cesar Chavez Day to National Farmworkers’ Day, as well as painting over murals of Chavez.

Miguel Santana, President at California Community Foundation, describes the dissonance he feels about these allegations in a piece co-written with his daughter, Amanda. Santana met Chavez in the late 1980s during a protest, and remembers him as “quiet, humble,” and “almost disarming in his gentleness.” Yet he writes with his daughter that accountability must look not like “protection of legacy, but alignment with values.”

And Rudy Gonzalez, a member of the San Francisco Labor Council’s executive committee, echoes the sentiment that this movement was never about just one man: “As a Mexican American labor leader, I was raised on the story of the farm worker movement — on sacrifice, on faith, on the belief that working people deserve dignity…but let me be clear: our movement has never been about one man,” he continued. “It has always been about workers — Filipino, Mexican, Black, immigrants standing together and demanding respect.”

It’s heartening to see these reactions from men, and also not all that common to see such an outpouring of solidarity. Women have been silenced since the dawn of time–with threats to their bodily and financial safety, self- and society-imposed martyrdom, and now non-disclosure agreements in exchange for money. Farmworkers, immigrants, and blue-collar workers are even more vulnerable to abuses of power–and perhaps this fundamental understanding is what united community leaders with the survivor community so vocally.

One silver lining is that Huerta was able to speak her experience into reality while she is still alive. I’ve seen stories take decades to come to light in a meaningful way–several of which, actually, are coming to the fore because of the Epstein Files. But it’s unusual for someone who is 96, as is Huerta, to be able to hold on to something so painful for six decades and then connect with others about it at the end of her life.

Today, I can’t think of a totally comparable larger-than-life figure in the labor movement to Chavez, perhaps with the exception of worker-focused politicians like Bernie Sanders, AOC, and Zohran Mamdani. And while I’m sure power can centralize in niche communities–and that people who do good things for the world can also be abusive to those closest to them–I am hopeful that labor movements recognize this is not ideal for the broader liberation movement.

In many ways, the revelations about Chavez remind me of the age-old conundrum of how people separate art from the artist. Can one separate a beautiful piece of music or film from an abusive bully who created it? With Cesar Chavez, we may have to hold several uncomfortable truths–all while ushering in a future that not only protects workers, immigrants, and farmworkers, but also refuses to silence future Dolores Huertas.

Share

What we’re reading…

A Northern Kentucky family declined a $26 million offer to build a data center on their 1,200 acres of farmland. One of the family members told local news: “If it’s my way, I’ll stay and hold and feed a nation. Twenty-six million doesn’t mean anything.” And: “I say they’re a liar, and the truth isn’t in them,” referring to claims the project would bring jobs and economic growth. “That’s what I say. It’s a scam.”

Mark Zuckerberg is building a personal CEO AI agent to help him do his job faster.

New York City has rolled out guidance for how schools and teachers should use AI. According to the guidelines, teachers can use AI to plan and draft documents, but not to assign grades or discipline students.

The freelance marketplace Fiverr is listing AI advertising and video directors “at the fraction of the cost” of traditional production teams. They’re targeting small- to medium-sized businesses who the advertising has historically been too expensive for.

Software developers are saying that vibe coding–which we wrote about last week–is “intoxicating.”

Senator Elizabeth Warren is launching a probe into how YouTuber Mr. Beast is promoting cryptocurrency to underage users. Beast Industries’ banking app Step previously claimed to be “the first financial app that will allow both teens under 18 and young adults to buy, sell, hold and receive crypto.”

A Substacker called DeepDelver wrote a scathing piece about the Y-Combinator startup Delve. They accused the startup of compliance fraud–and of outsourcing “automated audits” to a human audit firm in India. The founders of Delve are another addition to the list of “Forbes’ 30 under 30” founders accused of fraud.

The editor-in-chief of The Verge confronted the CEO of Superhuman (formerly Grammarly) about how the company impersonated him.

Here’s how the Iran war has thrown a wrench in the Gulf area’s concentration of energy AI power.