Anthropic Navigates White House's Improvised AI Export Rules

When the Commerce Department's Bureau of Industry and Security sent Anthropic a letter on June 12, 2026, directing the company to cut off all foreign-national access to its two newest AI models, it did so without a published standard and without disclosing the letter itself. That improvised quality is what Wired reported as central to this episode: officials told the outlet that Anthropic would need to guarantee its guardrails could not be jailbroken at all, a standard that most security experts consider technically impossible to meet. Anthropic put the gap plainly, stating that "to date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak."

The trigger was a claim that another company had successfully jailbroken Mythos 5, Anthropic's most capable model, which alarmed officials about national security risks. Anthropic disabled both Mythos 5 and its sister model Fable 5 for all customers while it worked toward compliance, stating publicly: "We disagree that the finding of a narrow potential jailbreak should be cause of recalling a commercial model deployed to hundreds of millions of people." A security researcher disputed the jailbreak characterization, describing it as defensive research instead.

The episode unfolded across a "whirlwind 24 hours" of tense calls involving Anthropic CEO Dario Amodei, National Cyber Director Sean Cairncross, Treasury Secretary Scott Bessent, and Commerce Secretary Howard Lutnick, according to Politico. Amazon CEO Andy Jassy reportedly shared the jailbreak claim with the government. David Sacks, the former White House AI "czar," said the administration viewed the issue as "serious" but one that "should be easily resolved."

It was, eventually. On June 26, the administration lifted its block, permitting Anthropic to release Mythos 5 to more than 100 US institutions, major companies and federal agencies listed in an attached annex. Anthropic committed to collaborate with the government on model protocols, standards, and release procedures. The sister model Fable 5 remains under review.

The full directive letter has not been made public, so what legal basis the order rested on remains unclear, and the reporting does not give you a settled answer on what jailbreak risk standard will govern future model releases or whether it applies to other AI companies. British MP Al Carns, a former minister for armed forces, captured the international reaction plainly: "This week the most advanced AI model on the planet got switched off by a foreign government." Allies are now reckoning with what dependence on US AI infrastructure actually means, and that reckoning is unlikely to fade as frontier models grow more capable.