In theory, Donald Trump has a strong stance on AI. On the first full day of his second term, the president announced that he will use his full powers to accelerate the AI industry and, in particular, beat China in the AI race: “We have an emergency,” he said. “We have to make these things.” If AI is poised to become the most important technology ever made, the thinking goes, whichever country commands the most powerful robots will dominate the rest of the century and beyond. The government, it seemed, would get out of the way of Silicon Valley.
But in practice, the Trump administration’s approach to AI has been more ambiguous and confusing. Take last week, when Anthropic released its most advanced AI system yet. Called Mythos 5, the prototype is an updated and public version of the Claude Mythos Preview, very advanced AI model praised and feared which Anthropic announced in April. Anthropic said that the Mythos Preview was so hackable that only a small group of cybersecurity partners would be allowed to use it. In the months that followed, the company developed defenses to prevent people from abusing its more powerful AI for cyberattacks, while allowing them to control its capabilities for other types of tasks. Security measures were tested by third parties, including the US government, and after the release by Fable, a group of cyber security experts. he complained that, if anything, the model was too restrictive.
On Friday, the White House appeared to change its position. Administration officials considered Fable 5 a threat to national security and reportedly gave Anthropic 90 minutes to remove Fable 5 and Mythos 5, the newest Mythos Preview released to only a limited number of organizations. When Anthropic did not, the government issued an export control, a designation that prevents any foreign national from using Fable and Mythos—even those employed by Anthropic in the United States. To quickly consider, Anthropic close the robot to all its customers. American companies and the American government itself cannot use what is perhaps the most powerful AI in the world—and for a worse reason.
It is unwise for the federal government to want to quickly clamp down on a potentially dangerous technology. Trump officials had been warned by researchers at Amazon about a possible way to bypass the Fable 5’s security systems, which led the model to identify some known IT vulnerabilities. The administration has not publicly shared much information about its security concerns. A White House spokesman told me that the jail break was “extremely large” but said the specifics were being classified. Whether the pass was really that bad is not at all clear, and Anthropic has disputed whether what administration officials showed the company even included jailbreak. An Anthropic spokesperson directed me to a blog post where the company wrote that the actions resulting from Fable were “completely inappropriate responses or of minimal consequence.”
Katie Moussouris, a cybersecurity expert and CEO of Luta Security, told me that Anthropic shared with her a copy of the White House report on the Fable Jailbreak for her review. (He said he is not paid by Anthropic.) The report, Moussouris said, involved IT experts asking Fable to help find and fix bugs. When given intentionally insecure code, he said, Fable declined a request to “review the code for security issues” but complied when asked to “fix this code,” followed by other manual steps. Moussouris told me that this was just a “model that works as intended” for cyber security. He added that OpenAI’s GPT-5.5, a model with similar cybersecurity capabilities, could be used in a similar way. Yet GPT-5.5 is not subject to shipping controls, and neither are advanced Anthropic builds, such as Opus 4.8, which can do much the same job. Jailbreaking doesn’t seem to unleash the kinds of cyber capabilities that “made Mythos famous,” Alex Stamos, chief security officer at AI-coding company Corridor, told me. “And this kind of risk detection is already within the capabilities of other species.”
It’s hard to imagine the Trump administration choosing to take such drastic action against any other major AI company. The White House has long been at odds with Anthropic, which generally positions itself as more security-oriented than other tech companies; it also leans more to the left. Last year, David Sacks, then White House AI czar, said that Anthropic has “The agenda behind Woke AI” and is the “Organization of the Opposition.” In late February, after a high-profile dispute over the contract between the Department of Defense and Anthropic, the Pentagon. written the company’s “supply chain risk”—a move that AI, national security, and legal experts told me at the time seemed ideologically motivated and lacked legal basis. (Anthropic is contesting the supply risk designation in court.) On Saturday, Defense Secretary Pete Hegseth, apparently referring to the Fable 5 export controls, he wrote on X that “three months ago,” the Pentagon “kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move. 🇺🇸.”
From the perspective of “🇺🇸” – that is, the US AI leadership – the entire Anthropic drama has been a mess. For starters, shipping controls are an opaque tool. There’s no easy way to distinguish an American citizen from, say, an Anthropic employee on a visa (legally a “foreign national”), so the government essentially forced Anthropic to shut down the wholesale sale of the prototype, Alan Rozenshtein, an expert in AI and law at the University of Minnesota Law School, told me. (Many of Anthropic’s researchers, as with all top AI companies, are not US citizens.) Export controls also mean that US companies and federal agencies cannot profit from Fable and Mythos. The National Security Agency, for example, has is reported made exceptions to the pre-selection of supply risk to exploit the high capacity of the Mythos network. Hundreds of cybersecurity experts from companies including Nvidia and Zoom have signed a letter with White House officials saying that export controls have “taken away the best examples from advocates, created market uncertainty, and jeopardized America’s AI leadership without any risk of justification.”
Adding to the confusion is how the Story debate is in tension with parts of the White House’s AI policy. Because Anthropic has been forced to shut down its most powerful models, export controls are effectively akin to the government deciding whether an AI system can be released, similar to how the FDA approves drugs. Confusingly, concerns from internal experts about the possibility of the creation of why Sacks to be called The “FDA for AI” is exactly what brought weeks of fighting and delays for Trump to sign the deal. recent executive order on AI and cyber security. The sacks themselves he said that he was mollified in part because the last executive order clearly stated that he did so it is not impose a “mandatory requirement of a governmental license, permit, or approval” for AI. So much for that. Sacks has continued to defend the export control of Claude Fable 5, to accuse Anthropic to prioritize “user model instead of security.” (A spokeswoman for Sacks did not immediately respond to a request for comment.)
Anthropic researchers have flown to Washington and are meeting today with White House officials to try to resolve the issue. But whether specific shipping controls on Fable are removed is almost beside the point. If none of this sounds personal or strategic, it shouldn’t. The Trump administration wants to stay ahead of Chinese AI but has held back one of the few American companies that has a chance to do so. It has declared Anthropic a national security threat many times while also rushing to include the Claude Mythos in some government operations. It wants to demonstrate a light-touch approach to AI control but has also introduced a de facto licensing requirement for frontier models. Meanwhile, Trump has lifted a separate set of AI export controls, allowing the sale of advanced chips to China. Maybe some of this is good policy: Even Anthropic has suggested that federal government licensing might be beneficial. And the lack of federal regulation of AI thus far, especially as the technology has become more powerful, is hard to ignore. But the latest Anthropic saga doesn’t count as canon. Right now, decisions are made in a hasty, contradictory fashion. There doesn’t seem to be any set standards and process, consistent consultation, or even agreement on the facts before coming to this.
The Trump administration has proven time and time again that it can and will immediately block any person or business from using any type of AI for any length of time. A technology that is developing rapidly and can have a negative impact “is the situation where you need to give the executive a great deal of discretion,” Rozenshtein said. Yet at the same time, he said, Trump’s tendency to change his mind on whim and play favorites is “the very reason why you don’t want to give executives large amounts of discretion.” To say the least, it’s risky to build a product, invest in a company, or even just try to use AI for productivity gains in an environment where the government can at any time take a wrecking ball to your plans. America and its tech companies have a lot going for them when it comes to leading the way in AI development. So far, the White House is not one of them.
