When a model can be switched off by email
The abstract fear of AI dependence became concrete on June 13, 2026, when a single US government directive forced Anthropic to take its two most capable models - Fable 5 and Mythos 5 - offline for users everywhere, including, briefly, its own foreign-born employees. The speed was startling; the international reaction, louder still.
Launch to lockdown in four days
Anthropic had made Fable 5 (the public face of its Mythos-class models) and the more restricted Mythos 5 generally available on June 9. Three days later, at 5:21pm ET on June 12, it received an export-control directive - issued by Commerce Secretary Howard Lutnick to CEO Dario Amodei - ordering it to suspend all access by any foreign national, inside or outside the US. Unable to filter users by nationality in real time, Anthropic said it had to abruptly disable access for everyone to comply.
The jailbreak at the centre
Washington's stated concern was national security, specifically a method for jailbreaking Fable 5's safety guardrails. Anthropic downplayed it, characterizing the technique as a limited ability to review code and spot errors - something rival models can also do. The government's account is sharper: David Sacks said the administration asked Amodei to fix the flaw or pull the model and that he refused. The Wall Street Journal reported the move was also shaped by Amazon's CEO, who told officials Amazon researchers had used Fable 5 prompts to obtain information that could aid cyberattacks.
A fight months in the making
None of this began last week. The dispute traces to Anthropic's insistence that its technology not be used for mass surveillance or fully autonomous weapons - a stance that angered the Pentagon and led the administration to bar federal agencies from using Anthropic's tools and to label the company a national-security supply-chain risk, a tag usually reserved for foreign-adversary firms. Anthropic sued to reverse the blacklisting.
The reaction heard around the world
Outside the US, the debate skipped past the jailbreak and landed on one uncomfortable fact: a tool embedded in companies, research bodies, and public services worldwide had been switched off by a foreign government in an afternoon. The European Commission said it was examining the fallout and warned contingency measures shouldn't discriminate against partners; a Finnish MEP said Europe can't keep relying on access a foreign government can revoke overnight; and Canada's Mark Carney framed it, heading into the G7, as a lesson in the dangers of over-reliance on a few American providers. The off-ramp runs through a Commerce Department licence - but for many capitals, the patch is beside the point. The lesson already drawn is that access to frontier AI is a question of whose jurisdiction holds the switch.