The US government has issued an export control directive suspending foreign access to Anthropic’s highly anticipated advanced AI models, Claude Mythos 5 and the more restricted version, Claude Fable 5. The order, delivered late on Friday, bars any foreign national, whether inside or outside the US, including Anthropic’s own foreign national employees, from using the company’s industry-leading cybersecurity-focused models.
Senior Anthropic technical staff are now reportedly in Washington, meeting with White House officials in a bid to resolve what has quickly become the AI company’s second significant dispute with the Trump administration in just over three months. Anthropic says it has disabled the models for all customers worldwide while it works out how to enforce the restrictions.
The episode carries high stakes not only for global AI users but for Anthropic itself, which is targeting a public listing valued at more than $1 trillion. That figure now hangs in the balance as the company navigates a second collision with the very government whose confidence it may need to go public.
A Jailbreak Claim, a 90-Minute Window, and a Standoff
The trigger for Friday’s order was reportedly a vulnerability in Fable 5’s safety guardrails. Amazon CEO Andy Jassy allegedly raised concerns with the White House on Thursday over the model’s potential to be “jailbroken.” A rapid sequence of calls followed on Friday between senior administration officials and Anthropic CEO Dario Amodei, during which Treasury Secretary Scott Bessent apparently told Amodei directly that he was making a “bad decision” by declining to voluntarily withdraw the model.
When Amodei declined to act, the administration moved swiftly. Anthropic says the letter arrived at 5:21 p.m. EST, citing national security concerns but offering no specific technical details.
Anthropic has contested the severity of the vulnerability, arguing that no testers have yet discovered a universal jailbreak and that the level of capability cited in the government’s report “is widely available from other models.” It also defended its systems, which it said “are so strong that many users have complained that they are overly broad.”
For many teams that had already integrated the models into their daily workflows, the move has been immediate. Robert Davitt, Product Lead at Fin, described the disruption bluntly:
“Feeling quasi-lobotomized working this morning without Mythos. It was a genuine redefinition of the frontier: elite intelligence, high agency, and the strongest reasoning I’ve witnessed. I’m looking at our team’s backlog and know Mythos would have blitzed a couple of high-priority tasks before lunch.”
A Pattern of Friction and a Telling Timeline
The export controls arrive just days after Fable 5’s general public release, which itself followed months of closely managed closed testing by a select group of companies and governments, including the US government itself. That timing has not gone unnoticed. The fact that Washington moved so swiftly after the model’s public launch, rather than during the extended test period when it had direct access, raises pointed questions about the motivations behind the order.
It is also not the first time the Trump administration and Anthropic have clashed. In February, Anthropic published a formal statement refusing to remove safeguards around mass domestic surveillance and fully autonomous weapons from its contracts with the Pentagon. The Department of War subsequently designated Anthropic a “supply chain risk,” a label the company called legally unsound, given that Claude remains embedded in critical national security operations across the US intelligence community.
That earlier dispute centered on principle; this one centers on capability. But both conflicts feed into the same underlying tension: the degree to which the US government can, and should, dictate the development, deployment, and protection of frontier AI models. The irony is sharp. Dario Amodei himself has long advocated for governmental oversight of AI, writing earlier this year: “I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries.” The fact that those words now sit alongside a direct government order against his own technology will not be lost on observers.
The wider geopolitical consequences are also drawing attention. Savinder Puri, Associate VP at Zensar Technologies, was among those to flag the implications for the global technology industry, noting that until this week, access to frontier AI had been broadly equal.
“Tech was democratically available: same cloud, same APIs, same models, anywhere on Earth,”
Puri wrote. “Today, that’s not the case.” Puri also raised specific concerns for India’s large IT services sector, questioning whether the directive represents a warning sign for an industry that has become deeply integrated with US-origin AI tooling.
What Comes Next
With Anthropic’s technical leadership in Washington and the company publicly calling the administration’s actions inconsistent with “principles of transparency, fairness and facts,” a resolution, if one is reached, is unlikely to be straightforward. The administration has signaled that it wants Anthropic to remediate the reported vulnerability so the export control can be lifted, but it has so far provided no timeframe.
For global users of Fable 5 and Mythos 5, the disruption is immediate and ongoing. Governments, including the European Commission and Canada, have flagged concerns, with Canadian Prime Minister Mark Carney warning that the incident highlights the risk of overreliance on a small number of powerful AI tools.
How Anthropic navigates the coming days will be watched closely not just by investors, but across the broader industry. Other frontier AI labs operating in heavily regulated sectors will be paying close attention to whether the company can satisfy Washington’s demands without compromising the safety principles on which it has staked its reputation and IPO story.
Both sides, for now, believe they are acting in the name of safety. They simply cannot agree on what that means. Whether this standoff can be resolved ahead of Anthropic’s listing window could determine not only the company’s valuation, but also how AI governance is managed going forward.
