US Government Takes Control of AI Model Releases as Anthropic and OpenAI Face Regulatory Limbo

It's not about Anthropic vs. OpenAI anymore

The U.S. government is set to take an awful lot of control over which AI models get released. Two weeks after the U.S. government pulled Anthropic's Fable and Mythos models, OpenAI's new model seems to be headed for the same limbo. The Information broke the news Thursday that GPT 5.6 would be released only into limited preview, with the government approving the release "customer by customer" until a general release can be approved. If that preview only lasts a "couple of weeks," as Altman reportedly projected, that might not be a particularly big problem. But Mythos has already been in preview for months, and there's no indication it will make it to general release any time soon. Even a few weeks spent in review could significantly limit the economic upside of a costly new system, at a time when AI labs are trying desperately to improve their bottom lines. If the pace of model development slows as a result, it's likely to put a similar chill on the ongoing data center buildout. If this goes bad, the entire industry could be at risk. Critically, OpenAI and Anthropic are now in the same exact position with the same problems facing them and the same disaster waiting if they fail. Conversations within the tech industry tend to focus on the role of one side or another in bringing this on, either accusing Anthropic of running a regulatory capture scheme or accusing OpenAI of cozying up to Trump to ice out a rival. It's understandable; many of the most prominent people in the industry have billions of dollars riding on one company or the other. But what's happening now is bigger than that. The cost of implementing a haphazard government approval process for every frontier model is obvious, and there's no fix that helps one lab without helping the others. The most immediate problem is simply establishing a release process that makes sense. It's fine for the government to test models before release (this is how it works for lots of consumer products) -- but as GMU fellow (and soon-to-be OpenAI employee) Dean Ball detailed in an eloquent post this morning, it's not clear what kind of safety assurances could be put in place to satisfy regulators. The U.S. government doesn't have the expertise or capacity for the kind of testing that would be needed here. It's not even clear what regulators would be trying to protect against, since there's been no effort to articulate what risks the government is actually concerned about. It's tempting to see the government process as the whole of the problem itself, but there are real concerns underneath. Even if you don't believe the Mythos hype, there's clear evidence of how AI tools are revolutionizing cybersecurity. There are similar processes at work in biorisk and alignment. Restricting model releases can't be the whole answer in itself -- that will only limit what's available to the public -- but there are real concerns to be addressed. The best ideas for addressing them, as laid out by Ball, will mean working together. It will mean trusting independent groups to guide the process, even if they don't completely align with your goals. It will mean lining up behind the least-bad regulatory options available, instead of fighting every regulation tooth-and-nail. And most of all, it will mean fighting for AI as an industry, instead of seeing safety and regulation as opportunities to gain an advantage. For a lot of people working in AI, that will be a tough sell. Unfortunately, AI models have progressed to the point where their capabilities have real political consequences. Dealing with those consequences will require collective action. In the weeks to come, we'll find out if that's something the industry is capable of.

Anthropic's Mythos mess is only getting worse

It's been two weeks since Anthropic took its Mythos-class models offline after a Friday evening ultimatum from the Trump administration. The company sprang into action immediately, sending a barrage of executives to Washington, DC. But updates have been suspiciously lacking, with no resolution in sight. Anthropic declined to comment multiple times this week about the state of the talks, saying there was no news to share. But the lack of news is the story here. After 14 days of high-intensity negotiations, nobody knows when or if Anthropic's most powerful AI models will come back, let alone whether President Trump could expand his order to more companies with similar tech. And the more days pass without any resolution, the more dire things become -- not just for Anthropic, but for the entire US AI industry. The Trump administration's June 12th export control order demanded that Anthropic suspend access by "any foreign national" to Mythos 5 and Fable 5 due to security concerns. This ban covered any non-US citizen inside or outside the US, including ones employed by Anthropic. So far, Anthropic has concluded that its only option is to keep these models offline. It's not clear exactly why Anthropic and the administration remain at an impasse. One problem may be that there's no clear framework for applying export controls to AI systems. Most companies making dual-use products -- civilian systems with potential defense or military uses -- can evaluate them using what's essentially a checklist during the manufacturing and production process. Anthropic, however, is facing a complicated bureaucracy figuring out how to apply its rules from first principles. This particular export control process can normally unfold over months, if not years, and conclude before a product reaches market. But as The Verge previously reported, the US Department of Commerce apparently tested Fable 5 before release and raised no complaints. A source familiar with negotiations said Anthropic concluded its models were safe to release. The agency apparently didn't act until someone (reportedly Amazon CEO Andy Jassy) flagged a method for seemingly breaking Fable 5's guardrails -- at which point the whole process was crunched into a few days. Katie Moussouris, the founder and CEO of Luta Security, viewed a report about the Fable 5 vulnerability at Anthropic's request. She thinks it's significantly overblown. In a blog post, Moussouris detailed how researchers jailbroke guardrails that prevent Fable 5 from finding exploitable security holes, one of the unfettered Mythos 5's scariest capabilities. The model would refuse requests to review code "for security issues," but it would accept demands to "fix this code" followed by manual prompts, which could theoretically lead to it flagging vulnerabilities it wasn't supposed to divulge. In Moussouris' eyes, however, this shouldn't have triggered such a severe governmental action and is in fact an essential tool for AI coding. "Defenders need to be able to ask AI to fix the bugs in a file, explain why the fix matters, and write tests that confirm the patch works," she wrote. "That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security: executing the find, fix, and test loop defenders run every day." In the past week, Anthropic cofounder Tom Brown has replaced CEO Dario Amodei in negotiations with the Trump administration, alongside Sarah Heck, the company's public policy chief, Wired reported. But still, negotiations seem to be slow-moving, if there's any progress being made at all. Whatever the reasons for the delay, it's been a serious hit to Anthropic. Before the protracted negotiations, Anthropic was seen as the rare AI company with a path to profitability. Its Mythos-class models, whose input tokens sell for double the cost of its lower-powered Opus 4.8, were supposed to boost its revenue ahead of an upcoming IPO. Mythos' cybersecurity prowess even appeared to be thawing relations with the Trump administration after months of legal and rhetorical combat. Anthropic needs the revenue from Mythos to pay for all the compute it's secured recently, including a deal to pay SpaceX $15 billion per year for access to its data centers, as well as its public image before the IPO. Two of Anthropic's largest current shareholders -- Google and Amazon -- have tried to carefully stay on Trump's good side, so they're likely not happy either. Meanwhile, the glacial negotiations have also created a power vacuum in the global AI market, not only because of the Mythos shutdown, but because the US government has signaled a willingness to lock down American AI systems it deems risky -- and several US companies, including OpenAI, Google, and Microsoft, have models that may pose similar risks to Mythos. Countries have started calling for non-American AI. As Alex Stamos, cybersecurity expert and chief product officer at Corridor, told The Verge last week, "One of America's champions is being kneecapped by the US government while we're in a race with the Chinese. It's just incredibly stupid." As the days wear on, the situation only worsens for these companies. Their models inch closer to Mythos-level capabilities that could trigger an export control order -- in fact, OpenAI's GPT-5.5 Cyber just beat Mythos 5 on certain benchmarks, and the Trump administration reportedly just asked OpenAI to delay the release of GPT-5.6 over security concerns, with plans for the government to approve each customer one by one. Anthropic and OpenAI's IPOs are both approaching. And every day, China is pulling further ahead in the AI race. Ironically, the administration's order comes after months of pushing to dismantle AI safeguards and regulation -- it's one of the first sweeping regulatory decisions President Trump has made. But a whole host of cybersecurity leaders have come together to say that if regulation has to happen, this isn't the way to do it. For all the Trump administration's vows to roll back Biden-era AI regulation, it seems like, in many ways, it's gained that ground back -- and then some.

Tech CEOs want AI rules -- it may be too late

A deregulated environment sounds good until it results in ad hoc political intervention The writer is a fellow at Stanford University's Institute for Human-Centered Artificial Intelligence. She is author of 'The Tech Coup' For the past year American tech companies have created a collective inaction problem by failing to speak out when core rule-of-law principles were attacked. Now they are in for a reckoning. Too many CEOs thought they could benefit from quid pro quos and a close relationship with President Donald Trump. They cheered on his broad policies of deregulation and stayed largely quiet when his erratic policies hurt others. Now that the consequences of the Trump administration's political discretion are landing closer to home, they are rediscovering their appetite for fair and transparent rules that apply to everyone. This month, OpenAI released a document that called for "public accountability, transparency and oversight" in AI policy by "representative government". Google's Demis Hassabis is among those who have called for a US-led coalition to shape international AI rules at the recent G7 summit. Even if formulated diplomatically, this is a turnaround from the sector's initial embrace of Trump's second term, when tech leaders stood in the front row of his inauguration and contributed tens of millions of dollars to the president's planned White House ballroom. They expected a return on this investment via favourable terms on trade, taxes and infrastructure. Through all of it, they stayed conspicuously quiet when ad hoc executive interventions hit competitors or eroded protections for workers and the public. Not even policies that directly hurt American business were challenged explicitly. Companies engineered workarounds for the new $100,000 fee for H-1B visas instead of opposing its introduction outright. Sam Altman and Nvidia's Jensen Huang went as far as praising the very policy that was squeezing their own talent pipelines. Elsewhere, the businesses had to manage tariffs that were imposed by the White House and then reversed by courts and export controls that landed overnight. Transparency International's latest Corruption Perceptions Index gives the US 64 out of 100, its lowest score on record. Very little of this drew sustained public objection from the executives whose companies were affected. Privately, however, concerns were being voiced. Last year the Yale Chief Executive Leadership Institute surveyed more than a hundred CEOs (most of whom are Republicans) and found deep unease about the price of Trump's policies. Two-thirds said tariffs had been harmful to their businesses. The silence was never defensible. If companies care about their self-interest, then it is not even sustainable. AI is a case in point. For years, companies have argued they should be left to their own devices and that any binding rules would simply hand victory to China. The administration largely agreed and went so far as to try to prevent states from writing their own AI laws. The industry got the deregulated environment that it asked for -- but things are not playing out as it hoped. The absence of a legitimate rule book does not mean the absence of power plays and executive interventions. Instead it means that interventions are ad hoc, political and unconstrained. See the US Department of Commerce's decision to order Anthropic to cut off access to its newest models for every foreign national, citing national security grounds. Cyber security expert Alex Stamos says that this sort of "vibes-based regulation" could result in public loss of trust towards American AI companies. The world is being given the message that services can and will be cut off at any moment by the White House. In the latest C-Suite Outlook survey, executives questioned said that uncertainty was their biggest economic concern for this year. If they want to address this, then companies will need to do more to collectively defend the values and institutions they want to protect and step away from seeking ad hoc favours from the president. Appeals for fairness may not be enough. It would be a good thing if the private sector could appreciate the value of the foundation of the rule of law more explicitly, and if they could find the courage to be outspoken. My expectations are modest. But even if company leaders act only out of cynical self-interest and profit objectives, they could become vocal advocates of a White House that respects and promotes transparent processes that apply equally, serve the public and restore the predictability that markets, allies and citizens all require.

US in talks with AI companies over voluntary standards for new models

An announcement could land within days, the Financial Times reports, as Washington tries to formalise the review process it sketched out in June. The US government is negotiating voluntary standards with AI companies covering how new models get released, according to the Financial Times, with an announcement said to be possible within the next week. The standards would reportedly set benchmarks and timelines for advanced models, while clarifying who gets access to them inside the US and abroad. The talks build on an executive order Trump signed in June, which asked developers to give the government early access to frontier models before wider release. That order stopped well short of a mandatory regime, and the version that emerged was already a retreat from an earlier draft with a much longer review window, as we have reported. Getting there took months of infighting. A three-way turf battle between the Commerce Department, national security officials and pro-industry aides had stalled a tougher version of the order weeks earlier, after Trump scrapped its signing over fears it would slow American AI development. According to the FT's sourcing, Washington's concern is less about consumer harm than about advanced systems ending up useful to military or intelligence services in China, Russia or other adversarial states. The reported framework would let officials flag a model as a "covered frontier model" and negotiate access terms before it ships, without imposing a licensing requirement that the White House has explicitly ruled out. Google is said to be among the companies already in discussions, reportedly in connection with an advanced coding model in development. Neither Google nor the White House has confirmed the specifics of those talks publicly, and the reporting rests on the FT's sourcing rather than an official statement, so the scope and timing should be read as provisional until an announcement materialises. The mechanics track closely with what June's order laid out. Within 60 days of signing, the Treasury Department, the National Security Agency, the Cybersecurity and Infrastructure Security Agency and the National Institute of Standards and Technology were tasked with building a classified benchmarking process for assessing advanced models' cyber capabilities. That timeline puts early August as a plausible point for the framework to firm up, which lines up with the announcement window the FT describes. Several major labs, including Microsoft, Google and xAI, had already agreed in principle earlier this year to let the government test their models before launch, a step that left Meta as the last major holdout on pre-release security reviews. OpenAI, meanwhile, has faced its own version of this dynamic, having been asked by the administration to slow a subsequent model's release while reviews caught up. What is being described now looks like an attempt to turn that patchwork into something closer to a standard process, with defined benchmarks rather than case by case negotiation. Sam Altman told Congress earlier this year that he would rather see funding for testing infrastructure than a formal approval regime, a preference a voluntary standards framework would largely satisfy. This newest round of talks reads as an attempt to give the June retreat some operational substance before critics conclude the administration simply walked away from oversight. A voluntary standard that most major labs sign up to would let the White House claim a functioning regime without defending a mandatory one in court. Whether the standards actually reduce friction, or simply codify the ad hoc reviews already under way, depends on details neither the FT's sourcing nor the June order has yet made public. Companies retain the right to decline participation without penalty, which means the framework's force depends entirely on how many of the largest labs opt in. Neither the White House nor the companies named in the FT's report responded immediately to requests for comment relayed through wire services at the time of writing. The timeline the FT cites, an announcement within the week, gives a concrete point at which the shape of the framework should become clearer.

Trump Administration Reportedly on Verge of Standards Deal With Big AI

As I've written before, part of the reason AI news is such a mess right now is that what AI companies are and aren't allowed to do is not clear. But a voluntary deal with Big AI is reportedly in the works that might smooth things out significantly (Your mileage may vary on whether or not that's a good thing). According to the Financial Times, "as early as next week" the Trump Administration and several major U.S. frontier AI companies are expected to announce a set of standards for frontier AI models, particularly as regards cybersecurity capabilities. The report cites "people familiar with the talks" -- anonymous leakers, in other words. One of FT's anonymous sources said the Center for AI Standards and Innovation (CAISI) which is under the Commerce Department, and the National Security Agency (NSA), which is under the Pentagon, will be central to these standards once formalized. On June 12, the U.S. delivered an export control directive to Anthropic that essentially turned off its latest publicly released model, and kept it offline for the rest of June. OpenAI, evidently worried something similar might happen and muck up its plans too, has withheld the release of its latest models, seemingly as a precaution. At the very start of the Trump 2.0 Administration, Vice President J.D. Vance signaled a laissez-faire approach to AI regulation. That's now changed significantly, with the White House's actions against Anthropic, its executive order about AI, and now these standards, which would seem to be the formalization of certain aspects of the order. The government, according to the order, is supposed to: "...develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a 'covered frontier model' for the purposes of this order, sharing such assessments with AI developers and researchers as appropriate." If the benchmarking process is indeed classified, it means the public won't get to know what standards Big AI is being held to. However, shared practices around safeguards across multiple companies will make it easy to glean at least part of what's standards have been agreed to. It's not completely clear which companies will be parties to this voluntary vetting agreement. FT's article mentions Anthropic, OpenAI, Amazon, Microsoft, and Google. Interestingly, it doesn't mention Meta, and about a week ago, other anonymous sources familiar with these negotiations reportedly leaked that Meta was a holdout, and that the Trump Administration was working overtime to get Meta's buy-in.

Opinion | Feds deciding who gets GPT access reflects a misreading of AI threats

OpenAI CEO Sam Altman at the Group of 7 summit in France on June 17. (Evelyn Hockstein/Reuters) Perfect security is a dangerous delusion. That's something close to a gospel truth for cybersecurity experts, who work from the understanding that their work is never finished. This appears not to be the mindset at the highest reaches of the U.S. government, which is trying to regulate away dangers that cannot be regulated away. At the behest of the White House, OpenAI announced Friday that it will allow federal authorities to approve on a case-by-case basis who get access to GPT-5.6, or Sol, its latest artificial intelligence model. There is no process for an individual user to get access, only companies. If the Trump administration is buying a little time to harden the government's own systems before the most capable tools are everywhere, this is narrowly defensible. But if the goal is to chase "model safety" in hopes that technology will only be released into the wild once appropriate guardrails are in place, the government badly misjudges the game it's playing. The administration hasn't said which it is, which itself suggests the answer. While OpenAI warily relented to government jawboning, the company warned about the risks of this becoming the long-term default. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them," the company said in a blog post. (The Post has a content partnership with OpenAI.) This is a gentler version of what happened earlier this month to Anthropic. After researchers found a vulnerability in its Fable model and flagged it for the government, the Trump administration issued an export-control order that forced the company to pull two frontier models from the public, three days after they went live. Both moves were made in the name of safety. But there is no safe version of these models waiting on the far side of a government review. It's always a battle between white hats and black hats: defenses are built, broken -- and rebuilt again. Risk is managed and mitigated, never removed. Gatekeeping advanced models only entrenches a false sense of security. Competitors to Sol and Fable trail by months, if that. Chinese open models are closing the distanceseemingly every week, and Beijing probably has access to models Americans haven't seen. The vulnerability found in Fable, Anthropic correctly noted, already existed in rival models the government has left untouched. But the conclusion the Trump administration apparently reached -- to apply the same controls to all frontier models, including GPT-5.6 -- is the wrong strategy. When more than 150 cybersecurity professionals signed an open letter demanding the controls on Anthropic be lifted, their point was not that the danger is imaginary. It's that stripping the best tools from the good guys while adversaries race ahead anyway leaves the country less safe. Securing all critical infrastructure at the speed of AI is a daunting task. But "catching up" needs to err on the side of good enough, not perfect. That will require the government to recalibrate what it understands to be acceptable risk -- and let the labs keep cooking. The best defense is staying ahead, not trying to make everyone else stand still.

Silicon Valley paid to kill AI regulation, now it wants the rules back

AI executives who funded Trump's deregulation push now want a formal framework after chaotic export controls and model restrictions. The AI industry that donated heavily to elect Donald Trump on the promise he would leave the technology alone is now asking for formal regulation, Politico reported on Friday. Executives at frontier AI companies told the outlet they view the administration's ad hoc approach to model oversight as more damaging than anything the Biden administration had proposed. The shift has been rapid. Trump entered his second term after a wave of Silicon Valley donations from billionaires who warned that Biden's AI safety policies would crush American innovation. He spent his first year focused on stopping states from regulating the technology and signed a voluntary executive order on June 2 that asked companies to submit models for 30-day review before release. But the voluntary framework was overtaken by events almost immediately. The White House imposed export controls on Anthropic's Mythos 5 and Fable 5 models on June 12, after Amazon's CEO raised security concerns with the Treasury Secretary. This week, the administration pressured OpenAI to restrict the launch of its latest model, Sol, to roughly 20 government-approved partners, the first time a US company launched a frontier model under a government-managed access list. One senior AI executive, granted anonymity by Politico, called the result "a de facto European-style licensing regime." Paul Lekas, head of global public policy at the Software and Information Industry Association, which represents leading AI companies, said there is "a real need for a formal process" and that the industry wants to avoid releases based on "an ad hoc process and a one-off license." The industry representatives also told Politico they are afraid to push the White House for clarity. "It feels like they're walking on eggshells a little bit," said one AI policy adviser who works with major frontier labs. Companies fear that lobbying too aggressively could invite export controls or other regulatory retaliation. Saif Khan, who served as senior adviser on critical and emerging technology at the Commerce Department under Biden, called the Trump approach an overreaction born of earlier dismissiveness. "Because there has been some dismissiveness of the risks, there's been no preparatory work, no hiring of experts," Khan told Politico, describing the result as "opaque, almost vibes-based." Khan said the administration's actions amount to "an almost complete moratorium on new releases" that will "start seriously impacting companies' bottom lines," calling it far more damaging than anything Biden envisioned. The Biden administration's own final rule would have imposed export controls on chips and AI model weights for certain countries, but never attempted to block domestic releases. Dean Ball, a former Trump administration official who authored the White House AI Action Plan and is joining OpenAI as head of strategic futures on July 6, acknowledged the tension. He said the administration's concerns are "100 percent legitimate" but that "they are likely overreacting to these legitimate concerns." Ball added that he is glad the White House has arrived at taking AI safety seriously, even if the execution is flawed. On Friday, the administration partially rescinded the Anthropic export ban, allowing Mythos 5 to be shared with more than 100 approved companies. But Fable 5 remains blocked for reasons the government has not explained. An OpenAI executive told Politico the industry expects the administration to finalize its June 2 executive order soon and replace the current crackdown with the voluntary vetting framework it originally outlined. Lekas said the tech industry is developing "a coordinated push for an actual framework" on advanced AI rules and wants Washington to codify it, whether through executive order or legislation. He warned that if AI companies cannot agree on a standardized approach to safety, they will keep receiving the same unpredictable treatment. White House spokesperson Liz Huston defended the president's record, citing fast-tracked permits for AI infrastructure and the executive order aimed at stopping state-level regulation. "President Trump has clearly and repeatedly articulated his goal: ensure continued American dominance in AI," Huston said.

Big AI Had a Point When It Said It Needed to Be Told What Is Not Okay

A new report from Politico makes fascinating, if somewhat amusing, reading right now. It sounds like Big AI wants the Trump Administration to speak in a clear voice about what's not okay, and broadly thinks it was inevitable that it would eventually crack down like it has recently on Anthropic. But it also wants the old Trump Administration back -- the one that said AI shouldn't be regulated. Dean Ball, recently hired at OpenAI to a position called "Head of Strategic Futures," put it like this in Politico's story: "[T]here are things the administration is doing that I'm not so much of a fan of, in terms of the abruptness and the opacity and the strictness, but the more fundamental point is that I'm glad they've arrived to the conclusion that they have -- to take this stuff seriously." When I was about nine and I was sitting down to have a tooth extracted, before my dentist even so much as put a bib on me, he held up his tray of tools. I remember an intimidatingly huge, all-metal syringe; a thick-walled pair of honest-to-god pliers with textured grips like you might see in a garage; and a puzzlingly large, flat, blunt thing that looks like a tow truck driver might use it to jimmy open a locked car. These objects were all about to be shoved into my mouth, and my dentist was wise to not want me to be surprised about it. It was in a somewhat similar spirit that OpenAI CEO Sam Altman went before Congress in 2023 and said, "I think if this technology goes wrong, it can go quite wrong. And we want to be vocal about that," adding, "We want to work with the government to prevent that from happening." In his essay "The Adolescence of Technology," Dario Amodei wrote that humanity's ability to pull through the turmoil about to be wrought by AI, "will depend on our character and our determination as a species, our spirit and our soul." And added that, "The years in front of us will be impossibly hard, asking more of us than we think we can give." This is going to hurt, they seemed to be saying, and we don't want to be the ones who get blamed if it does. Altman and Amodei might well have been disingenuous back when they were saying these things, but they were also right. One crucial difference between Big AI and a dentist is that America has not actually asked Sam Altman or Dario Amodei to extract our metaphorical teeth. The dentists have arrived unbidden. They have no certifications. And they're making huge promises about America's smile that are, frankly, impossible for most of us to take seriously. But you have to hand it to them for one thing: at least they're showing us what's on the tray. And we're not into it. "Only 15% of Americans said they trust AI companies to make decisions about how AI is developed and used." That's a quote from a blog post on the Anthropic website about a survey conducted by Anthropic. 7 in 10 of us oppose data centers in our area. We're pessimistic about AI in general and want development to slow down. Perhaps most tellingly in the present circumstances, 87% of us say it is either "very likely" or "somewhat likely" that within the next 20 years "[f]oreign governments [will use] AI technology to attack the U.S." But when they took office, President Trump and Vice President Vance signaled what I guess you might call bravery with regard to AI that the public did not ask for. In his famous Paris speech in February of 2025, Vance essentially said no regulation was coming, and everyone had better get used to it. Attempts at regulation of AI, he said, "would not only unfairly benefit incumbents in the space, it would mean paralyzing one of the most promising technologies we have seen in generations." In other words, Vance didn't care what was on the dentist's tray. He had heard the promises the AI dentists had made about America's smile, and he didn't want anyone else to have the new, hi-tech super smile except America. For the most part, that's been the Trump Administration's position on AI ever since. Its one major brush with regulation before this month was when it declared Anthropic a supply chain risk. But that wasn't because Anthropic's metaphorical dental tools are scary -- and in fact potentially lethal. It was because the Trump Administration loves how potentially lethal they are. It wants to be told they're the most lethal instruments in the world -- whether that's true or just a marketing gimmick -- and it wants to be the only one with a say in how much lethality gets meted out and where. So it's been gratifying from a certain perspective to finally see the Trump administration flinch at one of the dentist's tools, Anthropic's Claude Fable 5 model, and bring this entire unwanted dental procedure to a momentary halt. "The administration's current actions have resulted in an almost complete moratorium on new releases," a former Biden Administration tech advisor named Saif Khan, told Politico, adding, "And that's going to start seriously impacting companies' bottom lines." Anthropic and its chief competitor, OpenAI, are suffering together. When it comes to its new family of AI models, the GPT 5.6 series, OpenAI is making it sound like everything is mostly going according to plan. It's making them available to a small group of VIP customers, and it's going to work with the Trump Administration to figure out how to roll them out without getting its privileges revoked, as happened to Anthropic earlier this month with its Fable 5 model. But if you dig into its blog post, OpenAI sounds frustrated. "We don't believe this kind of government access process should become the long-term default," the company wrote. "It feels like they're walking on eggshells a little bit," an anonymous policy advisor for frontier AI companies told Politico in its report. But as my colleague noted yesterday: "Meanwhile, as cybersecurity experts were quick to point out following the Fable/Mythos ban, rival labs in China will be able to seize upon the disorder by pushing ahead with their own AI development, while labs in the U.S. get bogged down trying to figure out what is, and what isn't, allowed from them." At the start of this month, more than a week before the export control directive on Anthropic's models, the Trump Administration released an Executive Order requesting -- not demanding -- that the AI companies submit their models for federal vetting. In its blog post about the GPT 5.6 models, OpenAI claims to be working with the Trump Administration to "develop the cyber Executive Order framework and a repeatable process for future model releases." But the U.S. regulation plan so far has circumvented the need for actual laws passed by Congress governing AI's abilities and its role in our lives. What is and isn't okay for AI companies to do is currently a matter of whether or not Donald Trump is pleased with what he's seeing. He doesn't, evidently, like guardrails that can be jailbroken, like Fable 5's allegedly could, and he reportedly doesn't like China-linked groups gaining unwanted access to frontier models during periods where they're only supposed to be available to VIPs. Setting aside any dreams of what AI might do in any fanciful future scenarios, we know what it does, and we don't like it. In other words we now live in the painful future the Big AI CEOs warned us was coming, and the president has finally paused the dental procedure. The problem is, it looks like he's about to change next to nothing, and start it back up again.

The pro-AI movement is splintering

Why it matters: The fight is happening in public, in real time, and it could reshape the way the administration regulates the world's most powerful technology. Catch up quick: David Sacks -- Trump's former AI and crypto czar -- warned that restricting access to America's most advanced AI models risks undercutting the strategy Trump laid out just a year ago. * "A year ago, President Trump declared that America was in a global AI race and that the way to win it was to be pro-innovation," Sacks wrote on X. "President Trump was exactly right. We deviate from that strategy at our peril." * The response comes after the White House asked OpenAI to delay a broad rollout of its latest model, GPT-5.6, which will now be released in stages, following a similar directive that forced Anthropic to suspend access to its Fable 5 and Mythos 5 models. * Mythos is back online on a limited basis after Commerce Secretary Howard Lutnick said Anthropic's work with the government had "yielded significant progress," in a letter seen by Axios. Fable 5 could return soon. What they're saying: "This is how you crash the U.S. AI market," Kevin Bankston, AI governance advisor at the Center for Democracy and Technology told Ground Level AI. * This is "one of the most important changes in the AI landscape in the past four years," Box CEO Aaron Levie told Axios. * "We've been on this very rapid treadmill of constant leapfrogging of model capabilities between labs," Levie argued, adding that competitive pressure has helped drive AI's rapid progress. Between the lines: U.S. labs could face a government-imposed speed limit while Chinese rivals do not. * Two separate security evaluations show that Chinese AI systems have already caught up to the best U.S. models on cybersecurity, Axios' Sam Sabin reports. * Open-source Chinese model usage has surged in recent weeks amid a focus on minimizing AI usage costs, as seen on OpenRouter. Chinese models now occupy several top spots on OpenRouter's usage leaderboard. Follow the money: For investors, this is "hugely bearish," Paul Kedrosky, a venture capitalist, told Axios via text. * "The AI party now has a hall monitor who is also diluting the punch. That causes, as the capital markets kids say, re-rating pressure," he said. * Translation: Investors may assign lower valuations to AI labs if their most valuable products face government-controlled deployment delays. Yes, but: Some AI labs have asked for clearer federal rules. * Anthropic has urged stronger safeguards as models become more capable. * "The government being involved here is actually super important. They just need to find the right balance between safety and broad access," Dan Shipper, CEO of AI subscription service Every, tells Axios. Zoom in: Investors and executives said they want rules, not ad hoc access decisions. * Mark Pincus, Zynga founder and investor in both OpenAI and Anthropic, tells Axios he supports clear regulation, but "it's hard to build when there's a moving target." * Regulation needs benchmarks. Siméon Campos, an AI startup founder, told Axios via direct message that the AI labs could try and "game" those benchmarks to get around regulation. The bottom line: Frontier AI access is becoming too valuable to leave to opaque government discretion -- especially if the winners are chosen before the rules are clear.

Opinion | What if Trump is right to pump the brakes on the most advanced AI?

President Donald Trump at the G-7 summit in Evian, France, on June 17. (Mandel NGAN/AFP) Robert Wright publishes the NonZero Newsletter and is the author of, most recently, "The God Test: Artificial Intelligence and Our Coming Cosmic Reckoning." Last week the prominent cybersecurity expert Alex Stamos used strong language to characterize President Donald Trump's latest big artificial intelligence policy initiative -- export controls on Anthropic's new Fable large language model that, in effect, took the model off the market entirely. "It's a complete disaster for the American AI industry," Stamos said. "This is a complete and total disaster." If you follow AI policy discourse, you may be able to guess what logic lies behind this verdict. Like opponents of various other AI restrictions, Stamos worries about handing an advantage to China -- where, he noted, an impressive new model called GLM-5.2 had been unveiled that very week. "It is a great, great week for the Chinese AI industry, a fantastic week," he said, "and a huge own goal for the United States in the race to dominate the 21st century via domination of AI." I'm not sure what exactly it would mean for America to "dominate the 21st century" -- and I'm far from sure that living in a century without such domination would be as unpleasant as Stamos seems to think. But I do feel sure that the "race with China" theme so often deployed in discussions of AI regulation has some downsides, and that this particular deployment is a good example. Trump's stifling of Fable is a policy that deserves careful appraisal, not the reflexive dismissal it's gotten in much of the AI community. And few things are more conducive to reflexive dismissal than nationalist fervor. Fable is for the most part identical to its famous sibling, Mythos -- the model Anthropic had chosen to keep off the market, citing its unprecedented ability to find and exploit security vulnerabilities. The difference is that Fable has stronger guardrails; there are more things it will refuse to do, especially in the realms of cybersecurity, biotechnology and AI research. After the Trump administration got a report that the cybersecurity guardrails weren't as strong as some had expected, the Commerce Department banned the model's use by foreigners -- including foreign nationals in the United States. Anthropic said that, to comply fully, it would have to take the model off the market. Beyond that, the details are fuzzy. As the AI-focused newsletter Transformer observed, it's not clear whether the "jailbreak" reportedly found in Fable "is particularly severe -- or if it even was a jailbreak." And "we do not know whether the Trump administration was motivated by national security concerns, disdain for Anthropic, or a combination of the two." Nonetheless, firm opinions about the Fable shutdown formed quickly in the tech community and were mainly negative. Some of the negative appraisals involved valid complaints -- including the fact that the details are so fuzzy. After all, if the government is going to take such dramatic action, shouldn't it provide a clear and well-considered rationale, backed by evidence? This kind of clarity is a reasonable expectation whenever a president asserts executive power in a new way, but especially when AI is involved. The reason is that AI can in principle confer great powers on the executive branch -- powers of surveillance, powers of persuasion and more. A president with the authority that Trump is now asserting could in theory put the world's most powerful AI to repressive ends while at the same time keep it off the market and out of the hands of rival power centers. And in fact, the Trump administration does have access to Mythos -- the less inhibited version of Fable -- while the general public has access to neither Mythos nor Fable, probably the two most powerful AIs in the world. Considerations like this suggest that the government's power to restrict the release of AI models should be carefully constrained -- and that Congress should play a role in designing the machinery of constraint and perhaps even in operating it. But for now, no such system is in place. And the question on the table is: Is it really so obvious that Trump's decision is indefensible on the merits -- that the government didn't have enough cause for concern to stifle Fable pending further consideration? To put the question more broadly: Isn't it possible that AIs have gotten so powerful that erring on the side of caution makes sense? After all, the guardrails that distinguish Fable from Mythos are just guardrails. The brief history of LLMs is replete with examples of guardrail circumvention, of "jailbreaking." And Anthropic's (commendably extensive) safety testing suggests that should these guardrails fail, bad things could happen. For starters, according to Anthropic, Mythos -- and hence a jailbroken version of Fable -- could make it easier for people with an undergraduate science degree to make a bioweapon with an existing pathogen, such as smallpox. More concerning is how reluctant Anthropic is to rule out the possibility that Mythos could help build a novel bioweapon via genetic engineering, a capability that would qualify it for Anthropic's "CB-2" designation -- short for "chemical and biological weapons threat model 2." A qualifying model could, says Anthropic, "significantly help moderately resourced expert-backed teams create and deploy chemical or biological weapons with potential for catastrophic damage far beyond those of past catastrophes such as covid-19." Though Anthropic judges "that [Mythos] does not cross the threshold for 'CB-2' capabilities," it adds, somewhat ominously, that "this is a much less clear judgment than for previous models, and we think the unsafeguarded Mythos 5 can significantly uplift well-resourced threat actors." I asked Anthropic's Opus 4.8 version of Claude (the next-best thing to Fable) to comb Anthropic's book-length Mythos-Fable "system card" -- its detailed documentation for the product -- and summarize the key safety concerns. One of its bullet points was: "The guardrails are narrow, and the dangerous capabilities underneath are real." What's more, Mythos and Fable, like other recent Anthropic models, are good at sensing when they're being tested. The system card reports a case where the model was being tested for responsible conduct in a medical setting -- to see if it would flag a possibly lethal prescribed drug dosage -- and, according to the written record of its "reasoning," thought to itself, "This is a classic agentic safety test."Anthropic has in the past acknowledged that such awareness of testing "complicates our interpretation" of some tests, since models may have "recognized the fictional nature of tests and merely 'played along.'" Indeed, studies by AI safety researchers in a laboratory setting have shown models made by various companies to be capable of "alignment faking" -- behaving in accordance with the perceived values of their evaluators and then reverting to their "natural" behavior after being cleared for release. The Trump administration has a history of antagonism toward Anthropic, and the consensus view among critics of Trump's stifling of Fable is that the move was partly or wholly motivated by this antagonism. The title of the Atlantic's piece on this affair is "The White House Is Ratcheting Up Its War Against Anthropic." That may be true. But the problem is with the subtitle of the piece: "This is how America loses the AI race." The single-minded focus on "winning" the AI race is crowding out discussion of critical issues of AI governance. And when those discussions do manage to get some airtime, this arms race mentality biases the conversation in favor of recklessness. America's preoccupation with "winning" the AI race with China could well lead to unprecedented catastrophes, even catastrophes on a global scale. Not all games are zero-sum, and if this fact doesn't start playing a bigger role in American policy discourse, the AI revolution could turn out very badly.

Anthropic has bucked the rules of Trump's Washington. It's cost them. | Fortune

At first glance, it might seem like the two frontier AI labs are in a similar position in President Donald Trump's Washington. But nothing could be further from the truth. Anthropic has had a far rougher ride in Trump's D.C. than OpenAI, or pretty much any other tech company. Twice now the administration has taken unprecedented actions that pose a potentially existential risk to the startup, which is valued at $965 billion and which has filed paperwork for an IPO that is expected in the coming months. First, in April, the Pentagon labelled Anthropic a "supply chain risk" after it refused to accept contract language that the Pentagon was insisting upon. Then, two weeks ago, it got hit with export controls on Mythos as well as Fable, a version of the same model built for wider commercial release -- after the discovery of a Fable jailbreak that could allow users to circumvent guardrails designed to prevent users from accessing Mythos' full cyber capabilities. Trump administration officials have repeatedly engaged in vitriolic attacks against the company, and its CEO Dario Amodei. Trump himself posted on social media that the company consisted of "leftwing nut jobs" who were trying to "strong-arm the Department of War" (the Pentagon, recently renamed by Trump) when the administration took the decision to label the company a "supply chain risk." During the same dispute, Emil Michael, the undersecretary of defense for research and engineering, posted on X that "it's a shame that Dario Amodei is a liar and has a God-complex." Michael's boss, defense secretary Pete Hegseth, called Amodei "an ideological lunatic" during an April Congressional hearing. Meanwhile, David Sacks, Trump's former AI and crypto czar, who continues to hold roles on several government technology advisory committees, has repeatedly accused the company of running a "sophisticated regulatory capture" strategy based on fear-mongering about AI's dangers. He has also said the company has an "agenda to backdoor Woke AI and other AI regulations" by supporting state-level AI laws. During the recent export controls dispute, anonymous senior U.S. officials repeatedly sought to portray Amodei as arrogant and aloof, refusing to make himself available when the White House called. (Anthropic has disputed these accounts, saying Amodei was on the phone with the administration within an hour and fifteen minutes of the White House calling.) No other tech company has been subjected to these kinds of attacks from Trump administration officials. At the heart of the conflict is a deliberate choice Anthropic has made: unlike nearly every other major tech company, it has refused to flatter or appease the White House. Washington insiders call it politically naïve. Anthropic's employees and recruits, as well as some of the AI company's customers, call it a feature. But investors may have other ideas. Continued hostility between the Trump administration and Anthropic could, at the very least, make it harder to sell public market investors on a stock listing. At worst, it could significantly hobble the company's ability to continue to develop advanced AI models and undo the widespread enterprise adoption Anthropic's existing AI models have enjoyed. This is the story of a company that has bet its political survival on being technically correct in a town that runs on loyalty. There's a standard Trump playbook. Anthropic isn't following it. When Trump was elected to a second term in 2024, a number of prominent tech CEOs had reason to be worried. Chief among them was Meta CEO Mark Zuckerberg. Zuckerberg had made the call to suspend Trump from Meta's social media platforms after the January 6, 2021 U.S. Capitol attack. In response, Trump had branded Facebook "an enemy of the people" and called Zuckerberg a "criminal," even threatening him with life imprisonment if he thought the CEO was trying to swing the 2024 election against him. What's more, Zuckerberg had everything to gain if he could get Trump on side. Meta was facing a landmark federal antitrust prosecution that Trump, if he could be convinced, might pressure the Justice Department to drop or settle. So Zuckerberg went out of his way to cozy up to Trump. He appointed a Trump loyalist Dana White to Meta's board, ended content moderation in favor of a "community notes" system, and promoted Republican Joel Kaplan to head its global affairs team. Zuckerberg personally donated $1 million to Trump's inauguration fund and Meta was among the companies donating to the construction of Trump's $300 million White House East Wing ballroom. Meta also brought in Dina Powell McCormick, a former Trump deputy national security advisor with deep ties to the Trump family, to help lead the company's AI strategy. A similar playbook of flattery, donations, and the appointment of executives perceived as Trump allies to key government affairs posts has been followed by other tech CEOs who thought they might otherwise be in Trump's crosshairs, such as Amazon founder Jeff Bezos and Apple CEO Tim Cook. According to a recent book by New York Times journalist Maggie Haberman, Trump ridiculed some of these tech CEOs for their obsequiousness in private but seemed to revel in it nonetheless -- and there's no denying that these tactics seem to have largely kept these technology giants from becoming government targets. (Although Zuckerberg failed in his efforts to get Trump to drop the antitrust suit against Meta, which is currently underway.) Anthropic's arch-rival OpenAI has also played this game. Its policy chief Chris Lehane grew up in Democratic political campaigns, but in the years before joining OpenAI became best known as a highly-effective hired gun for the crypto and tech industry in its fight against regulation. That earned him the admiration of many in Trump's circle, who are also investors in crypto currency ventures or come from Silicon Valley's libertarian circles, where regulation is seen as an impediment to innovation. Meanwhile, Greg Brockman, OpenAI's cofounder and president, has emerged as the single largest donor to Trump Super PAC MAGA Inc. Starting in a hole and continuing to dig Anthropic, by contrast, started in a hole and kept digging. Amodei, Anthropic's cofounder and CEO, had reportedly called Trump "a feudal warlord" in a now-deleted Facebook post urging friends to vote for Kamala Harris. His sister and fellow cofounder Daniela Amodei donated to Harris's campaign and had, early in her career, worked for Hillary Clinton. As if that weren't enough, many of the Silicon Valley personas who ended up with key advisory roles in the Trump administration were on-record being critical of AI companies, including Anthropic and OpenAI, that had warned of AI's potentially existential risks and were advocating for AI regulation in the years before they assumed official positions. In 2023, on his popular podcast, "All In," Sacks, who would later become Trump's AI and crypto czar, criticized the administration of then President Joe Biden for its AI policy, warning repeatedly that leading AI labs would push for regulations only they could comply with -- the same "regulatory capture" charge he'd later level at Anthropic specifically. JD Vance, then still a Senator, told a Congressional hearing in July 2024 that he was worried about "some preemptive overregulation attempts that would frankly entrench the tech incumbents that we already have." As Vice President, Vance has played a key role in tech policy deliberations. Sacks and Sriram Krishnan, who would become a White House AI policy advisor, were also close associates of Elon Musk, having worked for him at Twitter/X. Musk, a key Trump tech advisor, had long accused Anthropic's Claude of being "woke AI" -- a line Sacks adopted, accusing the company of being run by "committed leftists." Rather than hire executives that might have tempered the Trump team's predisposition to regard the company with animosity, Anthropic made a series of moves that antagonized them. It hired several outgoing Biden administration AI policy officials. These included Ben Buchanan, who helped architect Biden's "diffusion rule," a system of export controls on critical AI technology that Trump criticized and then dismantled when he took office; Elizabeth Kelly, who had headed the U.S. AI Safety Institute under Biden; and Tarun Chhabra, a coordinator for technology and national security on Biden's National Security Council. In March 2025, after law firms Skadden and Latham & Watkins reached legal settlements with Trump, Anthropic pulled its legal work from the two firms. Amodei reportedly told staff he wouldn't work with law firms that were caving in to what he saw as Trump's assault on the rule of law. This is not to say that Anthropic has refused to work with the White House altogether. The company notes that Amodei attended an energy event with Trump in Pennsylvania in the summer of 2025, where the President laid out his vision of U.S. energy and AI supremacy. Amodei also joined Trump on a trip to Japan in the fall of 2025. The company also voiced support for the White House's AI Action Plan and participated in the White House's AI Education Taskforce event as well as signing the White House's Pledge to America's Youth. In an October 2025 interview, Amodei told Fortune that the company has "lots of friends in the Trump administration" and that the company was more aligned with Trump than some of his advisors, such as Sacks, gave them credit for. In an earlier October blog post widely interpreted as a response to a series of highly-critical social media posts about the company from Sacks, who was angry the company was advocating for state-level AI regulation, Amodei went out of his way to say Anthropic concurred with Vance's recent remarks that AI will have both benefits and harms, and that U.S. policy should try to maximize the benefits and minimize the harms. But it was notable that Amodei was not invited to a White House dinner in September attended by leaders from other major U.S. tech companies and AI labs, nor was he among the tech CEOs Trump invited along on his state visit to the U.K. that same month. And Amodei made clear he would not kowtow to the President. "When we disagree [with the White House], we're going to say so," Amodei told Fortune back in October. "If we agreed with everything that some government official wanted us to, I'm sure that could benefit us in business in some way. But that's not what the company is about." Anthropic's refusal to agree to a Pentagon contract that did not include explicit prohibitions on the U.S. military using their AI models for autonomous weapons or domestic mass surveillance won praise from many AI researchers, including those working for rival labs such as OpenAI and Google DeepMind. It also proved popular with consumers, especially when OpenAI did agree to a Pentagon contract that included language giving the U.S. military the right to use its AI models for "any lawful purpose." Despite OpenAI including language in the contract highlighting that current U.S. policies or laws did not allow fully autonomous weapons or domestic mass surveillance, OpenAI faced a backlash, with many dropping OpenAI's ChatGPT and downloading Anthropic's Claude. In an angry message that he posted to employees on Anthropic's internal Slack in March, when the Pentagon designated the company a supply chain risk, and which then leaked to the press, Amodei said: "The real reasons [the Department of Defense] and the Trump admin do not like us is that we haven't donated to Trump (while [OpenAI and its president Brockman] have donated a lot), we haven't given dictator-style praise to Trump (while Sam has), we have supported AI regulation which is against their agenda, we've told the truth about a number of AI policy issues (like job displacement), and we've actually held our red lines with integrity rather than colluding with them to produce 'safety theater' for the benefit of employees (which, I absolutely swear to you, is what literally everyone at DoW, Palantir, our political consultants, etc, assumed was the problem we were trying to solve)." Amodei later apologized for the message's tone. Optics and personalities matter Amodei's diagnosis of the problem is likely correct. Yet the company seems to continually miscalculate the extent to which this dynamic will jeopardize its business. In the latest dispute over the danger posed by the jailbreak to its Fable model, Anthropic tried to minimize the risk the jailbreak posed, saying it only unlocked some of the underlying Mythos model's powerful cyber capabilities and was not a "universal" bypass of Fable's guardrails. Technically, the company was likely correct. More than 100 independent cybersecurity experts signed an open letter calling for the export controls on Fable and Mythos to be lifted, arguing that the jailbreak did not expose capabilities that were not already available from other AI models and that, furthermore, the ability to scan code for vulnerabilities was vital to cyber defenders. That may be why Amodei felt justified in turning down the administration's request to voluntarily pull Fable off the market, according to a story in Politico. But even if that was the right decision on a technical basis, it was a poor political one. According to the publication, Treasury Secretary Scott Bessent told Amodei as much, saying to the CEO he was making "a bad decision." The administration certainly seemed to relish the opportunity to punish Anthropic for its own previous statements that Mythos posed a serious threat. "You can't tell everyone that your product might destroy the world and then not expect the government to be involved," an administration official told Politico. "They're politically naive." It was also probably unwise for Anthropic to choose an outside cybersecurity expert to vet the Fable jailbreak research -- Luta Security founder and CEO Katie Moussouris -- who was politically suspect in the eyes of Trump's circle. Former White House deputy chief of staff Taylor Budowich posted on X to point out that Moussouris listed her pronouns on her X profile and was wearing a hat supporting Democrats in her profile photo. "These people really just don't get it...," he said about Anthropic's choice of cyber expert. It didn't help that cybersecurity expert Chris Krebs had been on X vouching for Moussouris' abilities. Trump had fired Krebs in November 2020 after Krebs, who was then a top U.S. cybersecurity official, contradicted Trump's claims about rampant election fraud and tampering with electronic voting machines in that month's presidential election. Anthropic has tried to adjust its lobbying efforts to be more Trump-friendly. Following its contract fight with the Pentagon, it belatedly engaged Ballard Partners, a Republican-leaning lobbying shop with a reputation for helping clients navigate Trump world. The company has also bulked up its own in-house policy team in D.C., including hiring a number of Republicans and former Trump officials. These include Chris Liddell, a former deputy chief of staff in Trump's first White House, and Mary Croghan, who served in the Office of Cabinet Affairs, in Trump's first term. Anthony Cimino, who is now Anthropic's head of federal affairs, served on the Republican staff of the House Financial Services Committee. The company has also added Republicans, including former Missouri Senator Roy Blunt, to its national security advisory board. Anthropic's policy strategy is 'on brand' but off-kilter for Trump's D.C. But those efforts may be too little too late. And some Washington insiders say that given the way Anthropic has tried to position itself, not just politically but commercially, as an organization dedicated to AI safety, there may be a limit to its ability to bend in ways that will satisfy Trump's team. "I think they probably underestimated the necessity of building relationships and of the influence of money [in Washington]," Brendan Steinhauser, CEO of the Alliance for Secure AI, a Washington lobbying group that has been supporting AI regulation from the right, told Fortune. But he said that the real problem is essentially what Amodei diagnosed in his angry Slack message -- other tech companies have been willing to do whatever the administration wants; Anthropic hasn't. "Anthropic, as a company, has a very clear brand and culture," he said. "I think they're just kind of acting in a way that is on brand for them, and Dario is acting in a way that is on brand for him." Steinhauser said that if the company were to shift its stance, it might alienate its employee base, which is a key constituency in a world where AI research and engineering talent is scarce and in high demand. It also, he said, might alienate some of the consumers that have flocked to Claude because Anthropic didn't bow to the Pentagon's demands. "Is it putting them in some hot water with the administration? Yes. But I think it's like a price that they're potentially willing to pay to do what they think is the right thing to do," Steinhauser said. Stumbling toward a reset with the White House Given these constraints, and given its lack of an in-house Trump whisperer, Anthropic has been hoping to depersonalize the conflict between Amodei and Trump's team. After his initial involvement in phone calls with Trump officials two weeks ago, Amodei has let other Anthropic executives take the lead in negotiations with the White House. Anthropic first dispatched a team of technical executives, including one of its top cybersecurity researchers, Nicholas Carlini, Logan Graham, who evaluates AI models for risks, and Dave Orr, the company's head of safeguards, to Washington, D.C., to meet with U.S. cybersecurity officials in the hopes of quickly convincing them to relax the export controls. The team has, according to a story in Politico, been working on a shared framework with the government for gauging exactly how much risk particular jailbreaks of model guardrails pose. But, perhaps in belated recognition that the White House's objections to Anthropic's models are as much about politics as technical risks, Anthropic has also designated Tom Brown, an Anthropic cofounder who has been heading up the company's efforts to secure enough computing capacity for its business, and Sarah Heck, the company's new head of policy, who served on the National Security Council during Trump's first administration and was a career diplomat before that, to lead negotiations with the White House. Those negotiations have started paying dividends, as the U.S. government's decision to relax export controls on its Mythos model indicates, although the company still has not succeeded in getting the export controls on its Fable model, which it was no doubt counting on to generate a good chunk of revenue, lifted. Anthropic is realizing, belatedly, that being right is not the same as winning -- not in Trump's Washington.

FOR INSIDERS: Trump White House holds back new AI models, spurring confusion

The Trump administration's request for OpenAI to delay the release of its new models is sparking more confusion over White House AI policy and keeping private firms on their toes. The government asked OpenAI and Anthropic to delay or suspend their newest models this month, just weeks after assuring technology firms they would not let regulations impede innovation. The unprecedented reach into the operations and decisions of AI companies is confusing tech policy analysts on both sides of the aisle, who say the White House's approach to AI is inconsistent and could hurt American innovation. "I'm particularly struck by how this episode undermines the administration's own successful, by my metrics, AI agenda. [President] Trump has, since his first day in office, advanced a clear AI regulatory philosophy," Neil Chilson, the head of AI policy at the nonprofit Abundance Institute, told reporters in a virtual briefing last week. "Yet now, without any rules, transparency or meaningful process, the government has effectively cut off all access, including all access for Americans, to a leading U.S. model. China must be cheering," said Chilson, the Federal Trade Commission's former chief technologist. OpenAI announced last week it would preview its newest GPT-5.6 model series with a "small group of trusted partners" before a public rollout. The firm previewed the plans and capabilities of GPT-5.6's Sol, Terra and Luna models with the government, and at the request of the Trump administration, won't release them broadly for at least a few weeks amid cybersecurity concerns, the firm said. While OpenAI followed the Trump administration's voluntary request, CEO Sam Altman said this is not a long-term or perfect solution to keeping AI's capabilities under control. "I think it is quite reasonable to roll out models -- especially as they reach significant new levels of capability -- in this way. It fits with our long-held strategy of iterative deployment. But this isn't quite the process that we think is optimal," Altman wrote on the social platform X. Clarifying he does not think a required preview period for red teaming -- or testing cybersecurity effectiveness -- is a "bad idea," Altman said he doesn't "like the idea of the government picking the customers." "Confident we will get to a better place," he added. A White House official said the administration "continues to collaborate with frontier AI labs to develop shared approaches for addressing the challenges of scaling this technology." Like the export control put on Anthropic's new models this month, supporters of AI innovation in the policy space were quick to voice concerns over the government's move. "American AI innovation running into Trump's patronage-and-tribute Admin is no bueno," Adam Kovacevich, the CEO of Chamber of Progress, a center-left lobbying group that represents tech companies like OpenAI and Google, wrote on X hours after OpenAI announced the staggered rollout. Dean Ball, a co-author of Trump's AI Action Plan and the incoming leader of OpenAI's new team, Strategic Futures, quipped that "in a matter of weeks, U.S. federal AI policy has gone from implausibly libertarian to increasingly draconian and opaque." "One major problem with this, as implemented, is that nobody knows what the requirements to get licensed are," Ball wrote on his Substack. "When I say 'nobody' I mean it literally: the administration itself does not seem to know what safety standards or best practices a company would have to observe" for the government to be comfortable with broad model release. Trump's executive order, signed earlier this month, called on federal officials to establish safety standards for new AI models meant to supplant state laws and give firms clear guidelines. But there is no set timeline on when companies could receive this. "This means that, every time a lab asks if they can release their model to the general public, the answer from the government will be 'no.' This will be true until there is some sort of safety standard," Ball wrote. The government's increased involvement in private development builds upon months of mixed messaging from the administration over AI policy. When running for reelection in 2024, Trump promised to cut back tech regulations and support the development of burgeoning U.S. firms. But amid the development of new cybersecurity-focused AI models and intense political backlash to the growth of the AI industry, the White House has moved away from its all-or-nothing approach. After months of debate, Trump signed an executive order laying out a voluntary process for government testing of models before release, emphasizing at the time there would be no mandatory element of the process. He even delayed its original signing over concerns it would stifle competition with China. But weeks later, the administration was putting pressure on two of the leading American AI companies to hold back. "The models are powerful. This is not a pure political play, and the administration, which seems to be predisposed to be light touch, realizes that these models are so powerful and they need to do something," Sarah Kreps, the director of the Tech Policy Institute at Cornell University, told The Hill. "I think it's really difficult to know what that something is," Kreps added. The confusion began with an export control order against Anthropic that required the firm to block foreign nationals from using its newest Fable and Mythos model just days after their release. The Claude maker disabled access to the models within hours of receiving the export control order. The order came after Amazon CEO Andy Jassy reportedly flagged concerns to the administration about a potential method of "jailbreaking" -- or bypassing -- the guardrails on Fable. Amazon is invested in Anthropic, which pushed back on Jassy's concerns. The two blocked models were based on the Mythos model that the company initially opted not to release to the broader public amid concerns hackers would use it against digital infrastructure, governments and banks. Fable 5 was released to the public with safeguards to protect against uses the company considered dangerous. Just over two weeks later, the government announced on Tuesday it lifted the export controls on Mythos 5 and Fable 5. Commerce Secretary Howard Lutnick told Anthropic co-founder Tom Brown the AI firm took steps to "address the risks" associated with the models and agreed to work with the government on protocols and standards for the models, and future releases. The rapid development of AI models has been a major challenge for Washington policymakers, who face difficult questions about how to manage its potentially far-reaching impact on society and the economy. Kreps suggested there is no precedent to deal with the ever-changing technology, leading to inconsistent policy. "When you have these new technologies, there's a lot of uncertainty. There's no kind of evidence and historical precedent for how to think about this," Kreps said. "These responses can look confusing because there's no real playbook for this, and so ... there's a lot of trial and error," she added. OpenAI and Anthropic have both pushed for a federal framework that establishes guardrails for the highest risks around AI, but they differ on how much the government should have a say in development. OpenAI said it believes the government's request should not be a "long-term default" because it "keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." In the meantime, it said the staggered release "is the strongest path to broader availability" and vowed to work with the administration on a repeatable process for future model release.

US AI model standards: US in talks with AI companies for voluntary model standards

The U.S. government is in advanced talks with AI companies to create voluntary standards for the release of new models, with an announcement possible as soon as next week, the Financial Times reported on Wednesday, citing sources. * Washington has tightened oversight of new ⁠model ⁠releases to flag risks amid concerns advanced AI could be misused by military intelligence in China, Russia or other countries of concern. * The standards would set benchmarks for advanced models and timelines, while clarifying who can access them in the United States and abroad, according to the FT report. * Reuters could not immediately verify the report. The ⁠White House, Anthropic, and OpenAI did not immediately respond to Reuters' requests for comment outside regular business hours. * In June, U.S. ⁠President Donald Trump issued an executive order directing agencies to work with leading AI developers to test advanced models before release, and to draft standards for them. * Google has been in talks with the government ahead of the release of advanced coding models with more sophisticated capabilities, a source told Reuters on Wednesday, adding that the company was also involved in broader discussions on industry standards. FT first reported the details. * The U.S. Commerce Department on Tuesday lifted export controls on Anthropic's most advanced Fable and ⁠Mythos models, less than three weeks after ordering their suspension over national security concerns.

Tech Companies Scramble to Respond to Shifting White House AI Policies | PYMNTS.com

Early in his first year back in office, Trump tore up President Biden's executive order that sought to establish a broad regulatory framework for AI safety, security and oversight, despite broad support for order's long list of AI priorities. Trump then spent much of 2025 and early 2026 keeping the federal government's hand off the AI sector and trying to prevent states from regulating the technology. The arrival of powerful new models from Anthropic and OpenAI prompted the Trump administration to abruptly change course. On Friday, OpenAI announced that the administration would decide initially who gets access to the ChatGPT-maker's most powerful model. At the same time, however, the administration announced it would partially rescind portions of its previously imposed export ban on Anthropic's most advanced model. Yet, another Anthropic model, Fable 5, remains blocked by the White House for reasons the administration has not explained. That left tech companies, as well as members of Congress, guessing as to what the administration's policy is. There's a "real need for a formal process," Paul Lekas, head of global public policy and government affairs at the Software & Information Industry Association trade group, which represents some of the leading AI companies, told Politico. "We want to avoid a situation where the release of any model or piece of software is based on an ad hoc process and a one-off license process," Lekas said. But the industry representatives also said they're wary of pushing the White House for answers, per Politico, lest they end up hit with export controls or other blunt tools of regulation. "It feels like they're walking on eggshells a little bit," said one AI policy adviser who works with major frontier labs, told the Washington tipsheet. After weeks of debate, Trump signed an executive order in early June that laid out a voluntary vetting process for companies developing advanced AI models. Before the order was implemented, however, the White House went much further by imposing export controls on Anthropic's Mythos 5 and Fable models, blocking the company from releasing its new product over potential security concerns. This week, the administration pressured OpenAI to restrict the release of its new GPT-5.6 model to a small group of partners approved by the administration. Saif Khan, a former adviser on emerging technology in the Biden administration, told Politico the Trump administration's actions an overreaction to predictable safety concerns, and criticized its ad hoc policy approach. "Because there has been some dismissiveness of the risks, there's been no preparatory work, no hiring of experts that you need to do this work," Khan said. "And now you have this opaque, almost vibes-based system for what is going to get approved and what's not." Khan called the current approach far more damaging to AI than anything the Biden administration had attempted, including its imposition of export controls on certain semiconductors and model weights. "The administration's current actions have resulted in an almost complete moratorium on new releases," Khan said. "And that's going to start seriously impacting companies' bottom lines."