The Trump administration has issued an unprecedented export control directive forcing Anthropic to suspend global access to its most advanced AI models, Claude Fable 5 and Mythos 5. The order, which applies to all foreign nationals anywhere in the world—including non-US citizen Anthropic employees—has effectively crippled the company’s latest offerings and raised serious questions about the future of American AI leadership .
Anthropic received the directive on a Friday evening with only 90 minutes to comply, according to company statements . The Commerce Department, led by Secretary Howard W. Lutnick, informed Anthropic CEO Dario Amodei that a license is now required for any export, reexport, or transfer of the models to any destination worldwide and to any “foreign person,” wherever located . Failure to comply would result in “prompt criminal and civil penalties” .
The government’s stated concern centers on a potential “jailbreak”—a method to bypass safety guardrails built into Fable 5 that could allow full access to Mythos 5’s advanced cybersecurity capabilities . The administration believes Fable 5’s protections can be disabled, allegedly making it a national security risk . However, Anthropic has strongly disputed the justification, arguing that the government has only provided “verbal evidence of a potential narrow, non-universal jailbreak” .
The company emphasized that the vulnerabilities identified appear “relatively simple” and that “other publicly-available models are able to discover them as well without requiring a bypass” . Anthropic warned that “if this standard was applied across the industry, it would essentially halt all new model deployments for all frontier model providers” .
Since Mythos debuted in April, Anthropic has consistently warned about the model’s dual-use nature. The company itself acknowledged that “a great deal of advanced usage of AI models is dual use: the same queries that are beneficial in the hands of cybersecurity professionals and biology researchers could be dangerous if available to malicious actors” . Claude Mythos is capable of identifying and exploiting zero-day vulnerabilities across major operating systems and web browsers, even at the request of users without technical expertise .
Experts, however, argue that this crackdown obscures a far more uncomfortable reality. Tarah Wheeler, chief security officer of TPO Group, called the restriction “myopic in the extreme,” stating that “there are other companies hot on Anthropic’s heels who probably have the capabilities, too, and are holding them in reserve as they see how Anthropic is being treated in the current regulatory environment” .
Anthropic itself has emphasized this point since the launch of Mythos Preview. Logan Graham, the company’s frontier red team lead, told WIRED: “The real message is that this is not about the model or Anthropic. We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months” . Many of the assumptions that we’ve built modern security paradigms on might break,” Graham added .
Bruce Schneier, a researcher at Harvard University and the University of Toronto, echoed this sentiment, noting that “smaller, cheaper, open-source models, sometimes by themselves and sometimes in concert with each other, can match Mythos/Fable’s performance with more sophisticated prompting. And we should expect other models to match Mythos/Fable’s creativity and tenaciousness within months—slightly longer for open-source models” .
The directive comes amid an already strained relationship between Anthropic and the Trump administration. Earlier this year, the Department of War designated Anthropic a “supply chain threat,” effectively severing business ties after the company refused to allow its AI models to be used for domestic mass surveillance or the development of lethal autonomous weapons systems . President Trump has reportedly labeled Anthropic “radical left” and its executives “leftwing nutjobs” .
The timing is particularly critical for Anthropic, which has filed a draft prospectus for an initial public offering. Both Anthropic and OpenAI are reportedly valued at over $1 trillion, and the administration’s actions could jeopardize the capital pipelines these companies depend on to continue competing in AI development . The directive also creates a de facto licensing scheme that could reshape and potentially down-skill America’s AI workforce, as many key Anthropic personnel were born outside the United States .
Project Glasswing, an industry consortium convened by Anthropic, had already been working to address these very concerns. The group, including Microsoft, Apple, Google, Amazon Web Services, and more than 40 other organizations, had private access to Mythos Preview specifically to give developers time to mitigate vulnerabilities and exploit chains . The idea was to draw on tenets of coordinated vulnerability disclosure—giving developers time to patch bugs before public discussion .
The broader implications for American AI leadership are now coming into sharp focus. If the US government can cut off access to the latest models without notice, building those models into companies’ and governments’ IT infrastructure creates unacceptable risks . The message to the rest of the world, as one analyst put it, is that while the US might lead the world in AI development, it isn’t a reliable partner . China’s open-source AI products could start to look more appealing .
The debate is no longer about whether the technology carries risks, but whether specific restrictions meaningfully reduce those risks or simply slow down those trying to improve security . The net effect, as Anthropic argues, is that “a lone operator can now turn a month’s worth of patches into working exploits in a single afternoon” —a reality that no export control can reverse .
Anthropic remains in talks with the White House to restore access to the models but has yet to secure an agreement . The company continues to argue that no model yet exists that doesn’t have potential jailbreaks, and that competitors like OpenAI and Google have similar capabilities and vulnerabilities . As the situation unfolds, the fundamental question remains: are we truly addressing the risks of advanced AI, or are we simply delaying the inevitable?
Discussion about this post