Hello and welcome to Regulator, an email for Verge subscribers about technology, politics, and what happens when science crashes headlong into self-interest. Not a subscriber? Sign up here today! Got the scoop on a petty feud that’s going to somehow fundamentally reshape the entire field of frontier AI development? Send ’em over to tina.nguyen+tips@theverge.com.

Back when I was covering Donald Trump’s first presidency, it was incredibly common to read three different versions of the same story. His administration had split into several factions, all of which had different interests, and all of which hated each other. There was the Reince Priebus traditional GOP faction, the Manhattan society-based Jared Kushner faction, the proto-populist Steve Bannon faction, the deep state John Kelly faction, the conspiracy-MAGA Mike Lindell faction, and so forth. Over time, you could get a sense of which camp was leaking which narratives to the media, either to undermine their rivals or to save their own reputations. In fact, for several decades, media manipulation was a common survival tactic in Trumpworld, which often ran on factionalism and fierce competition for Trump’s approval. As The Associated Press reported on November 26th, 2016, in an article about his pre-presidency business management style, “Aides also often float suggestions to him through the media, knowing that Trump is a voracious watcher of cable TV and might be persuaded by what he sees and hears.”

This tendency toward back-biting and cross-purposes once again gives us a little insight into the White House, this time explaining some of the omnishambles around the Trump admin’s Friday night decision to impose licensing restrictions on Anthropic’s advanced Fable model. Even after several days of statements, reports, and negotiations, it’s still unclear what actually happened behind the scenes, and it’s even less clear who’s responsible for what could have been a massive cybersecurity disaster.

To summarize for the uninitiated: On Friday evening, the White House placed an export control restriction on Anthropic’s Mythos 5 and Fable 5, its most recently released models, preventing foreign governments and nationals from using those two products. It essentially forced Anthropic to completely shut off access to both of those models, throwing their user base into chaos, the company’s future into doubt, and the overall prospects of frontier AI into a gray zone. Could the government simply just tell AI companies to stop operating? Over the weekend, several conflicting narratives emerged, though a few points remained consistent: The White House’s allies claim that within days of launch, several tech executives, including Amazon president and CEO Andy Jassy, reached out with concerns that Fable and Mythos could be jailbroken, posing an imminent threat to national cybersecurity. The two sides spoke on Friday, though the nature and duration of their calls vary, depending on the account: The Washington Post’s account claimed that Anthropic was given 90 minutes to take down its models, while a White House official told Politico that they had begged Anthropic “for hours.” 

And from there, the details become even foggier, at least from the POV of the White House. One party tells The New York Times that Amazon had found a way to “jailbreak” the safety guardrails preventing users from using Fable for cyberattacks. A second party countered to the Times that one could achieve the same results with OpenAI’s ChatGPT 5.5. Semafor reported that it may have been due to a China-linked group accessing Mythos, though no actual jailbreak had been confirmed. And Axios’ sources indicated the admin simply did not like Anthropic’s woke vibe. “Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences,” one source familiar with the administration’s thinking told Axios. “It’s like they just speak in different languages.”

The Verge senior AI reporter Hayden Field has a more thorough report here on the view from Anthropic’s world and the AI safety community, which argues that the fears around the jailbreaks are overblown. But inside Washington, the overwhelming consensus I’m hearing is that even if they’re ultimately right, Anthropic and Dario Amodei have created their own political nightmare by opposing the Trump administration — and not in a morally superior #resistance way. “They don’t bend the knee and Dario is stubborn and says what he thinks even when it’s dumb and they have a (justified or not) holier than thou vibe,” one AI policy advocate noted to me.

One of the most interesting themes emerging in Regulator’s run is how often the hard logic and irrefutable facts inherent to science and technology run headlong into the vibes-based venality of politics. And in the absence of any federal regulatory law, AI regulation is fully reliant on the vibes of whatever statement someone’s convinced Donald Trump to post on Truth Social or executive order he’s been talked into signing. Unfortunately, as former White House AI adviser Dean Ball pointed out on his Substack on Tuesday, this is the reality that frontier AI companies must operate under. As he writes in a blistering post:

What the law says does not matter. What Administration officials argue on one day does not matter. Anthropic is a political enemy of this Administration, in part because they have explicitly chosen to make themselves one. It is simply naïve to think that your company can operate under such circumstances without an extreme degree of regulatory caution. And given this context, Anthropic’s actions are viewed by many within Washington as not simply unwise, but actively antagonistic.

And it is not just about Anthropic and political grudge matches with the Trump Administration. Everyone at the frontier should understand that in practice, you do need an explicit green light from the government now.

That mantra certainly applies to all companies operating under the Trump administration, particularly to companies working on emerging technologies. But for some reason, while its competitors have quickly adapted, Anthropic’s vibes have been more visibly out of sync with those of the administration — and, tellingly, there are no factions in the administration rushing to defend Anthropic. Which has got to raise some sort of question.

And now, Recess.

Shot: Two weeks ago, before the New York Knicks and the San Antonio Spurs face off in the NBA finals, Sesame Street‘s Elmo betrays his hometown.

Chaser: The Knicks win in five, but the fan base hasn’t forgotten.

See you next week.

For months, Big Tech’s Washington lobbyists have chased after the holy grail of pro-AI legislation: preemption. Their dream would be a comprehensive federal law, passed in Congress and signed by the president, applying one set of AI rules across the entire country and overriding the legally messy state-by-state approach to regulation. For months, lobbyists have run into roadblocks and incurred nationwide political blowback, and they now face the possibility that after the midterms, Congress will flip to hostile Democrats unwilling to work with them.

But their final, most desperate attempt at preemption is coming with new baggage, related to an entirely different fight in Congress that predates the public launch of ChatGPT: child safety. 

Earlier this week, reports leaked that the White House had told child safety groups and Big Tech companies that it would endorse a slate of children’s online safety laws backed by Sen. Marsha Blackburn (R-TN), the coauthor of the Kids Online Safety Act (KOSA), as part of an overall preemption package. While the issue of online safety does meaningfully overlap with AI, it’s only one facet of a much larger, complex set of issues that would need to be addressed in a truly comprehensive law: frontier model safety, discrimination, environmental impact, and so forth.

Regardless, the potential deal has hit one snag: The White House had apparently not informed House Republicans, which had just passed their own version of KOSA out of the , that it was going with Blackburn’s legislation as a vehicle. Democrats who’d worked with Blackburn on the Senate’s flavor of KOSA were allegedly left out of the loop, too. On top of that, there was a separate, bipartisan-backed AI preemption bill currently floating around the House.

It resulted in a week of total confusion for backers of either policy: AI preemption and child safety might be lumped together in order to ensure preemption gets signed into law, but whose version of child safety gets passed is unclear. Was it the Senate’s stricter KOSA? Was it the looser version backed by House Majority Leader Steve Scalise (R-LA)? And where was the White House in all of this?

“No one knows really who’s driving this thing,” a Republican lobbyist for a midsize tech company told The Verge. “Everyone is deeply, deeply, deeply skeptical of [the bill’s] movement, because everyone is on such different pages. I think the House is not going to move anything that Blackburn wants.”

Though the AI regulatory fight has caused huge fissures between GOP leadership and their populist members, President Donald Trump himself has called for the passage of an AI preemption bill, meaning that the Republican Party must somehow make this happen. These days, the White House’s policy wonks are trying to finesse a preemption approach influenced by Mike Davis, a Trump-allied lawyer and the founder of the Article III Project, who led a successful attempt to kill a different AI moratorium in the Senate last year.

Broadly speaking, to win Davis’ approval, preemption law should meaningfully protect a set of values Davis called the “Four Cs”: children, conservatives, creators, and communities. Some of those values were included in the White House’s proposed draft of a comprehensive AI law, released in March of this year, and the inclusion of KOSA satisfied the “children” requirement. But Davis told The Verge that he wanted any legislation to address all four. “There is no chance in hell AI preemption will pass if it does not address the Four Cs. I will make damn sure of that. Again.” 

Getting KOSA passed, however, involves reconciling a massive difference between the House and Senate versions of the same bill. The Senate’s version would require tech companies to assume a “duty of care,” preemptive measures to protect young users, and would extend that responsibility to AI companies as well. However, the House version, spearheaded by Scalise, diluted that provision late last November, to the fury of child safety advocates. The House’s exclusion from the White House’s discussions, therefore, was notable to onlookers. “[Blackburn] genuinely does not want House KOSA,” noted Michael Toscano, senior fellow and director of the Family First Technology Initiative for the conservative Institute for Family Studies. 

Even if Trump were to whip the House Republicans into line, they’d have another problem: the congressional Democrats, who’d also learned of Blackburn’s negotiations with the White House at the same time the House Republicans did. Though Senate KOSA was cosponsored by Sen. Richard Blumenthal (D-CT) and overwhelmingly passed 91–3 in 2024, they had not been aware that their legislation would now be handcuffed to the unpopular goal of AI preemption. “If they [Blackburn and the White House] are looking at a standalone bill, it’ll have to go through the Senate,” said an AI policy advocate, noting that a new version of this bill would then require 60 votes — and therefore, Democrats — to pass.

And even if the bill had some amount of popularity, the schedule might not allow for it. “It is mid June. You have a month and a half before people leave for [five-week] recess. And then it’s [general] election season,” said the AI policy advocate. “There’s just no way.” The remaining weeks on the legislative calendar are already being sucked up by more immediate matters: the renewal of FISA, an immigration crackdown package, increased defense spending for Trump’s war with Iran, a crypto market structure bill, affordability measures, and the controversial SAVE America election bill. Oh, and the regular budget items like Medicaid.

Having preemption and KOSA chained together presents Big Tech with a difficult choice: Do they want federal AI preemption more than they want immunity from “duty of care”? And they don’t have much time to make this choice, noted the Republican tech lobbyist, especially if Democrats take one chamber. “After the election, what incentive do the Democrats have to support anything? Like, why wouldn’t they say, ‘Fuck you, we’re gonna do our thing in the new Congress?’ I’m deeply skeptical.” 

Austin Carson, the former head of Nvidia’s government relations operations and the founder of SeedAI, a nonprofit focused on increasing access to AI for local communities, was more dubious that the KOSA-preemption shotgun marriage of convenience would succeed. “I can’t imagine a scenario where [this bill] would move,” he told The Verge. “I cannot imagine it.” 

Correction, June 17th: A previous version of this post characterized preemption as a blanket and mischaracterized the status of the House Republican KOSA bill; this has been updated to reflect that there are different scopes across the various preemption bills and that the House Republican version of KOSA passed the House Energy and Commerce Committee, not the House.

Hello and welcome to Regulator, a newsletter for Verge subscribers about tech politics, tech influence, and tech shenanigans in Washington, DC. (If you’re not a subscriber, you can get on board here.) We’re back after a two-week hiatus, during most of which I was gallivanting in the Netherlands for a family wedding, and a trip to the Heineken Experience, which is, truly, an ~experience~.

Before I left, I asked everyone in Washington to please chill out while I was gone. This clearly did not happen, and I have returned to a political landscape that can be best described as that meme from Community where the room is on fire. Let’s get into that.

If you wanted to get a good sense of how Washington insiders viewed the release of Magnifica Humanitas, Pope Leo XIV’s encyclical laying out Catholic doctrine on artificial intelligence, let me take you inside an actual room of Washington insiders.

Here’s the scene: a black-tie gala last week at the Waldorf Astoria, which used to be the Trump hotel, held by the Washington AI Network. In attendance, spotted among the dancers dressed like robots on stilts: AI lobbyists, AI safety nonprofits, tech industry representatives, tech journalists, Shark Tank’s Kevin O’Leary, senior administration officials — Administrator of the Centers for Medicare & Medicaid Services Mehmet Oz, Department of Energy Under Secretary Darío Gil — and me. Papal nuncio Archbishop Gabriele Caccia, the Vatican’s top diplomat to the United States, is also there, making his surprise debut to deliver remarks to the assembled, who expected to celebrate the breakout power players of artificial intelligence. (Yeah, Kevin O’Leary was receiving an award. It was that broad of a celebration.)

The nuncio is trying to relay the pope’s message of safeguarding humans and the human condition before innovation and profit. But I can barely hear him. The salad course has come out, and Caccia is being drowned out by the sound of cutlery on plates and people murmuring to their tablemates, because this is prime networking time.

Even if the general public is excited about Magnifica Humanitas, the pope doesn’t carry the force of law, nor can he impose onerous regulations, and therefore the pope does not immediately matter to Washington. For all the dinner talk chatter, the AI industry seems to be experiencing some tunnel vision. Generally, corporate lobbyists try to befriend everybody, Democrat and Republican alike, and cultivate those relationships for as many years as possible without pissing off either side. But that’s not a possibility in Donald Trump’s Washington, where supporting a Democrat in the past could be viewed as disloyalty, even for tech oligarchs. (Billionaire and commercial astronaut Jared Isaacman’s nomination for NASA administrator, for instance, was iced for several months after Trump learned he’d once donated to a Democrat.) 

On the other hand, if they give him money and make him look good, Trump can be convinced to give those oligarchs every regulatory ask they want and force the Republicans to do what he says. But even that control is tenuous. Here’s the incredibly brief recap of Trump’s most recent AI-focused executive order:

• On May 20th, several outlets report that President Trump will imminently sign an executive order that establishes a government review board for unreleased advanced AI models, delaying release by a maximum of 90 days.
• On May 21st, Trump decides not to sign that executive order, thanks to last-minute lobbying from David Sacks and Elon Musk.
• Trump then reverses this decision and signs that order after all on June 2nd, thanks to even more last-minute lobbying from Treasury Secretary Scott Bessent, but changes the timeline to a 30-day maximum.

But although Washington can be chaotic and unpredictable, especially when Trump is president, there are two fixed points in time that everyone can plan around: once every two years, a federal election will take place in November, and the winners of those races will be sworn into Congress the following January. There will, inevitably, be some change in the balance of power. But no one can safely anticipate who will hold that power or what it will look like, leading to an infinite series of unknowns for tech companies: What happens if the Republicans lose the House? What happens if they lose the House majority by one member, or 10, or 20? What does this scenario look like, but in the Senate? Which Democrats will take control of what committees? What if Alex Bores gets elected? What if a Trump loyalist pushes out a Republican ally of ours? What happens if a friendly Democrat gets pushed out by a progressive for something we have no control over, like their support of Israel? And so forth. 

But the tech industry’s own interests may be a key issue in the upcoming midterms. It’s easy for a voter to grasp the consequences of famous, instantly recognizable Big Tech CEOs standing behind Trump during the inauguration, or a gold statue from Tim Cook changing the anticipated price of an iPhone, or a check from a tech giant funding a gaudy ballroom (of all things). In this cycle, it’s even easier for voters to draw a straight line from those visible moments to the increasing and often unwanted presence of AI in every aspect of their everyday lives. Those voters complain to their representatives, those representatives respond, and if they don’t, the voters push them out of office in November.

A related light reading recommendation: One of the resources I’ll certainly be drawing on a lot during that time is journalist Molly White’s new project, Tech Influence Watch, which is tracking all the AI industry political spending in the coming midterms. White had begun the project as a way to track crypto spending in the 2024 election, but in the post announcing its expansion, pointed out that crypto and AI politics were inextricably linked — so linked, in fact, that the donors and strategists driving the AI super PACs were the exact same people. “The PACs may look different from the outside, but they’re increasingly the same operation with aligned goals: deregulate the tech sector, slash consumer protections, and allow tech companies to capture even more enormous profits at the expense of everyday people.”

And now, Recess.

See you next week.

By the time that the Democratic primary for New York’s 12th congressional district wraps up in June, Anthropic and OpenAI will have spent millions on their battle over the political future of AI: who gets to regulate it, or who will be punished for trying to regulate it. But the real winner of their feud may be the guy they’re currently fighting over: a once-obscure New York state assemblyman, who they’ve Streisand-effected into becoming the poster child for AI safety regulation. 

Ever since late 2025, Leading the Future, a super PAC funded by OpenAI, Palantir, and a16z executives, has spent millions against Alex Bores, who wrote one of the first pieces of AI regulatory legislation in the country. The PAC hoped to kill his bid for the seat about to be vacated by longtime Democrat Rep. Jerry Nadler. Instead, Bores is now a front-runner in the eight-person race to become the “face of Manhattan,” as New York Magazine recently put it in a cover feature. 

And shockingly, he pulled all this off without running a massive ad campaign. In fact, the Bores campaign told The Verge that it had placed its very first ad buy in New York on May 11th, nearly seven months after Bores entered the race and only weeks before the polls close on June 23rd. In contrast, Leading the Future, whose backers include Joe Lonsdale, Marc Andreessen, and OpenAI’s Greg Brockman, has been running attack ads against Bores since December 2025, spending an estimated total of $2.4 million according to the most recent reports.

In any other situation, a corporate- and billionaire-backed super PAC, which is allowed to raise and spend unlimited amounts of money for a candidate or ballot initiative so long as it does not coordinate with the campaign, could outspend a target into oblivion. The group behind Leading the Future had already done so successfully in the 2024 election, ousting Sen. Sherrod Brown (D-OH) and Rep. Katie Porter (D-CA) via Fairshake, a crypto industry super PAC. And it would surely have the upper hand in Manhattan: Way back in December, Think Big PAC, which is affiliated with Leading the Future, spent $120,000 to run a single anti-Bores attack ad on television and digital. “It is so expensive to advertise in a New York primary,” said Lis Smith, a New York-based political operative who ran Pete Buttigieg’s dark horse 2020 presidential campaign, and whose career involved stints with Govs. Andrew Cuomo and Eliot Spitzer. “The New York media market is the most expensive media market in the country. You’d kill for any bit of air time.” 

When Bores entered the race last October, the ex-Palantir employee-turned-public servant faced several other candidates with broader name recognition and deep-pocketed backers. Micah Lasher, a fellow New York state assemblyman, has the backing of Nadler’s New York political machine, as well as Mike Bloomberg’s super PAC. Jack Schlossberg, the influencer and grandson of John F. Kennedy, has the support of the national Democratic establishment nostalgic for Camelot. George Conway, the Donald Trump critic and ex-husband of former Trump campaign manager Kellyanne Conway, has won over the Never Trumper crowd. “I’m gonna be honest with you, [Bores] wasn’t exactly a well-known quantity prior to becoming a target of these AI companies,” Smith said. 

Reached for comment, Josh Vlasto, a spokesperson for Leading the Future, said, “From day one, we have said what is now playing out in plain sight: Alex Bores is bought and sold by Anthropic, its investors like Chris Larsen, the Effective Altruist community, and a network of dark money fringe tech groups. He has three super PACs funded by Anthropic, its investors, executives, and their allies who are trying to buy regulatory control of AI for themselves. Any group or endorser who thinks that by backing Alex Bores they are taking on billionaires is either foolish or needs to do a quick Google search.” 

Instead of smothering Bores’ candidacy, however, the AI companies only elevated his visibility. A new poll released last week by Emerson College shows him neck-and-neck with Lasher, trailing him by two points, and Bores has consistently taken the lead against his rivals in recent polling. In an odd way, Leading the Future’s ad buys became an in-kind donation to the Bores campaign, in the form of a free, multimillion-dollar ad campaign that did the hard work of raising voter awareness that Bores even existed.

Alyssa Cass, a spokeswoman for the Bores campaign, said that they’d initially thought it would be harder to make voters care about AI safety. “AI [regulation] is his strength, but it’s gonna take us a lot of work to make this a salient issue in the district,” she told The Verge. “And they, starting in December, started doing that work for us, of raising the saliency of AI and AI regulation, and making people think: Who are these people? What do they want to do to me, and to our society? For people for whom it wasn’t top of mind, they made it top of mind.”

Suddenly, voters who’d been unaware of Bores were getting mailers and ads describing him as anti-AI and pro-regulation, highlighting his authorship of the RAISE Act, a New York state law that placed restrictions on the release of frontier AI models and had been signed into law in December. 

The ads became an unintentional signal booster: The more that Leading the Future attacked Bores, the more media coverage it drew, and the more voters became aware that a super PAC backed by Silicon Valley AI billionaires was trying to influence a Manhattan election, specifically targeting a candidate who wanted to regulate AI. It also inadvertently gave Bores the “it” factor that differentiated him from seven other Democrat candidates running in Manhattan on a platform of holding Donald Trump accountable: He was, quite literally, being targeted by tech billionaires who were donating to Trump’s ballroom in exchange for AI deregulation. Internal campaign polling, in fact, shows that voters who had received negative information about Bores were more inclined to vote for him. 

It also didn’t hurt that the AI companies’ ads were easy for Bores to mock. One of their first ads criticized Bores for working at Palantir during a period where the controversial tech company had contracted with Immigration and Customs Enforcement, and suggested that Bores was hypocritical for now saying he wanted to abolish ICE. Bores filed a cease-and-desist against LTF for defaming him, claiming that he had left Palantir because he opposed its relationship with ICE. He also noted on social media that it was “ironic” that a Palantir billionaire was attacking him for working at Palantir. 

And this was even before an intervention from Anthropic brought the race to national awareness. In February, the Jobs and Democracy PAC, which is affiliated with the pro-regulation super PAC Public First Action, announced that it would back Bores. Outlets like The New Yorker, The New York Times, and Politico began covering the fight over Bores as part of the long-standing rivalry between OpenAI and Anthropic, the AI company that positions itself as the more responsible of the frontier labs — and which had just donated $20 million to Public First Action, in direct opposition to Leading the Future and its ties to OpenAI. 

This kind of media exposure is unheard of in House races. “You’d kill for any earned media, you’d kill for any paid media,” said Smith. “So the fact that he’s getting all this paid media, when he was a virtual unknown outside of extremely political insider circles before — it’s a gift.” 

Hello and welcome to Regulator, a newsletter exclusively for Verge subscribers about tech, politics, and Washington intrigue. (It’s basically House of Cards, but for nerds.) Not a subscriber yet? You really should become one, and to save you a Google search, here is the direct link to do so! And do you think I should know something? Send it to tina.nguyen+tips@theverge.com.

On Monday, The New York Times reported that the White House was considering having the government review AI models before release. To the casual Verge reader, it appeared to be a total reversal in Donald Trump’s policies. For the past year, he had been a vocal champion of pro-industry deregulation, repealing former President Joe Biden’s massive executive order on AI safety, lifting export controls on advanced chips, and signing executive orders that would have legally punished states for passing and enforcing AI laws in the vacuum of federal legislation. Now, the Trump administration has seemingly pulled a 180, demanding federal oversight and vetting of pre-market models.

But to Washington, the shift in the White House’s policy was due to three major changes. First, Anthropic’s Mythos has genuinely spooked the national security apparatus, forcing the administration to confront a new threat: the possibility of adversaries using American AI models to attack America’s public and private sectors. Second, other countries are now beginning to lay out their own AI regulations, potentially in a manner that would go against the interests of the United States. (And yes, “destroying a Big Tech data center in a targeted drone strike” is a manner of government AI regulation, but we’ll get to that shortly.)

And third, David Sacks was pushed out of his job as the AI and crypto czar, giving Silicon Valley one less mechanism to pitch an industry-friendly, “innovation-at-all-costs” agenda to Trump himself.

The definition of political influence can be squishy and amorphous, especially around Donald Trump, who will pick up anyone’s calls and then act on that advice if he feels like it. (Remember when Laura Loomer had control over the National Security Council?) But what’s legally certain is that Sacks, the billionaire venture capitalist and Trump fundraiser in 2024, no longer has the privileges available to him as a special government employee, such as the ability to review sensitive information, to speak on behalf of the White House, or to hold official influence over government employees and agencies. 

Instead, the “special government employee,” who was supposed to only spend 130 days working in the administration and somehow stuck around for an entire year, actively undermined the administration and torched its relationship with its political allies. During Sacks’ tenure, the White House went beyond simply advocating for less regulation. They tried twice to get Congress to pass a moratorium on state AI laws, and failing that, tried to use an executive order that would grant the Trump admin the powers to sue states passing or enforcing said laws. But his Valley-esque tactics, to say nothing of his attempts to consolidate power over AI policy by boxing out existing agencies, ended up infuriating Republican and MAGA allies, while alienating vast swaths of Trump’s base. (In fact, it was so unsuccessful that when unnamed White House officials recently attempted to pressure certain red states into dropping pending AI legislation, claiming that they were going against Trump’s agenda, four GOP state lawmakers spoke on the record to The Wall Street Journal instead. Then again, if the agenda was just to kill those bills in the cradle, it succeeded.)

Even if Sacks hadn’t crashed and burned — to say nothing of publicly criticizing Donald Trump, a man who does not like being criticized, for continuing to wage war against Iran — the job is also getting harder for one part-time employee maintaining ties to the private sector to handle. In recent months, the aperture of America’s AI policy has widened to a scale much, much broader than Sacks’ pro-innovation 2025 remit, into areas where a lack of regulation would be wildly irresponsible: national security and geopolitical stability.

A major turning point was the leak of Anthropic’s Mythos, the AI model that was so powerful at finding cybersecurity vulnerabilities that the company, whose reputation hinges on acting more responsibly than its competitors, refused to release it to the public. The possibility of a Mythos-level model becoming commercially available spooked the national security apparatus and the financial industry, and seized the attention of three powerful White House figures: Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and Chief of Staff Susie Wiles. 

When Bessent and Wiles met with Anthropic CEO Dario Amodei in April, it signaled that not only were they taking the threat seriously, they were now overriding Anthropic’s enemies in the Pentagon, who had, months prior, convinced Trump that Anthropic was “woke” and should be banned for government use. 

“The national security implications of something like Mythos are hard to deny, and legitimately urgent national security issues are not easy to politicize,” Charlie Bullock, a senior research fellow at the Institute for Law and AI, told The Verge. “Once serious national security people get involved, it’s hard to dismiss or politicize the issue.”

In recent weeks, federal agencies that had been boxed out by Sacks are now being given more authority. On Tuesday, the Commerce Department announced that it had designated the Center for AI Standards and Innovation (CAISI) as the agency that would conduct pre-deployment testing on commercial frontier AI models before release, and had already struck agreements with xAI, Microsoft, and Google DeepMind. CAISI is run by the National Institute of Standards and Technology (NIST), which had been gutted by Elon Musk’s Department of Government Efficiency (DOGE) last year, but has begun hiring for technical positions. 

Other countries are flexing their muscles, too, in a way that the United States can’t directly control. The European Union is currently debating a revision to the AI Act, and though the EU countries and the European Parliament were unable to reach an agreement in the most recent negotiations, whatever does end up coming out of that legislation will have a direct impact on how frontier AI models are developed — and possibly in a way that inadvertently acts against America’s business and natsec interests.

“[Bessent] really doesn’t like the Europeans,” a tech policy adviser close to the administration told The Verge. In his view, the EU’s proposed privacy regulations would not just hurt American companies, it would also inadvertently allow China to develop faster, and there was historical precedent: “We’ve seen this movie before when it came to broadband, where they tried to do the same thing to American broadband companies. In the end, they were just helping Huawei.”

Then there are rogue geopolitical players who simply do not care what Sacks or the American government think. Days after US forces bombed Tehran and killed its religious leader, Iran conducted drone strikes on two AWS data centers in the United Arab Emirates and indirectly damaged a third data center in Bahrain, causing major power outages across the Middle East and damaging critical infrastructure. Weeks later, Iranian state media announced that it would directly target 18 major US tech companies with a presence in the region, including AI heavy hitters like Google, Meta, Microsoft, Palantir, and Nvidia, and has since claimed that it hit an Oracle data center in the UAE. (UAE media later clarified that an Oracle building in Dubai suffered minor damage from falling debris from an aerial drone interception.) 

“While there’s a lot of politics around it here in the United States, where Maine is [trying] banning them, and [Republican Florida Gov. Ron] DeSantis is talking about banning them, as far as the world is concerned, this is critical infrastructure,” the tech policy adviser close to the administration told me. (DeSantis has proposed empowering towns to reject data center proposals and banning their construction near schools, but has not publicly suggested banning them entirely.) “And that’s why one of the first things the Iranians did was bomb not just one, but two of Amazon’s critical data centers, because they know how important it is.” The damage to AWS’s data centers, which service the entire Middle East, are severe enough that even if the war were to end now, it would take “several months” to resume full operations, according to the company.

This is not to say, however, that David Sacks has no influence in the Trump administration: He has Trump’s direct cell, and he’s still a billionaire CEO, which is a better credential in Trump’s eyes than any kind of expertise. The Atlantic’s George Packer recently published a massive feature on Sacks, highlighting his plutocratic tenacity. One can’t really stop a master of the universe from trying to flex every so often. But even when it comes to Trump’s favorite rich guys, Sacks may not even stack up. Last week, Trump hosted a state banquet for King Charles III’s visit to the United States. The guest list included Tim Cook, Jensen Huang, Jeff Bezos, Marc Andreessen, Marc Benioff, corporate leadership from Meta and Alphabet — and no Sacks, who had been present at a previous state banquet held at Windsor Castle in the United Kingdom last year, while he was still at the White House. 

When I asked a DC insider familiar with state dinner politics if Sacks had been invited, the answer was pretty telling: “Why would he be? He’s not White House inner circle.” 

And now, Recess.

 See you next week.

 Update, May 7th: Clarified Ron DeSantis’ position on data centers.