AINEWS
Anthropic disabled its two newest AI models for every user, including Americans
A US government directive citing national security forced the global pull. Anthropic says the technical claim behind it is narrow and disputed.
A US government directive citing national security forced the global pull. Anthropic says the technical claim behind it is narrow and disputed.
Anthropic pulled its two newest AI models from every user on Friday, including Americans on US soil and the company's own foreign-national staff, after a US government directive citing national-security authority forced the move. According to a ZDNET report by David Gewirtz, the disable is the first widely visible case of an American national-security tool reaching inside US borders to take a frontier AI model offline, with the underlying technical justification publicly disputed by the company that built it.
The directive requires the lab to block access to Claude Fable 5 and Claude Mythos 5 by "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees," as Anthropic relayed it. Rather than carve out a foreign-national-only restriction, Anthropic disabled both models for all users worldwide, effective 5:12 p.m. ET on Friday June 12, 2026. Other Claude models remain available. The disable is scoped to the two newest releases.
Anthropic framed the technical premise as a "specific technique" that identified "a small number of previously known, minor vulnerabilities", which is the company's own characterization, not a neutral fact. The government, according to Anthropic, asserted that the technique constitutes a method of "bypassing, or jailbreaking" Fable 5 and presents a national-security risk. The government has not publicly disclosed what authority it invoked, what the technique is, or how long the disable will last. That disclosure asymmetry is now the substance of the dispute.
The structural break is the scope. The directive, as Anthropic described it, applies to US-based users and to Anthropic's own US-based foreign-national staff, not just overseas users. That is not classic export control. It is a domestic disable of a frontier AI tool for everyone, ordered under a national-security rationale the public cannot yet evaluate. Whether the technical claim is real or narrow is a separate question, and the kind the government has, so far, declined to answer in public.
A prior ZDNET report documented Claude Fable 5 throttling AI researchers, which set up public attention to the model line and may help explain why a directive aimed at Fable 5 has now produced a global outage rather than a quiet patch. Anthropic's choice to disable the models for Americans too, rather than attempt a foreign-national carve-out inside its own US operations, signals that the directive did not leave room for a partial fix.
What is still missing is the part a reader would need to weigh the dispute. The legal authority behind the directive is not named. The specific jailbreak technique is not described. The duration of the disable is not stated. And there is no independent technical assessment yet of whether the vulnerabilities Anthropic calls "narrow and minor" are in fact narrow and minor. The next concrete fact to watch is whether the US government publishes any underlying directive text or technical explanation, and whether Anthropic follows the global disable with a narrower, model-specific patch once it can.