US bans Anthropic's Fable 5 and Mythos 5 for all non-citizens, citing jailbreak risk
Anthropic disabled access to its two most advanced AI models, Fable 5 and Mythos 5, after the Trump administration issued an export control directive on June 12, 2026, barring any foreign national—including Anthropic's own non-citizen employees—from using them. The order cited national security concerns and alleged discovery of a bypass to the models' safeguards.
The jailbreak in question was narrow and technical: security researchers (reportedly from Amazon) discovered that asking Fable 5 to "fix this code" rather than "review code for security issues" could unlock its ability to identify software vulnerabilities. Anthropic disagreed with the ban, stating the capability is not unique to Fable and that the jailbreak cannot be fixed without crippling the model's usefulness to defensive security teams.
The move comes amid escalating tension between Anthropic and the Pentagon. The company has refused Pentagon demands to allow its models for autonomous weapons and mass domestic surveillance, leading the Defense Department to bar military use of Claude in March 2026. Anthropic is now negotiating with Commerce Department officials in Washington to restore access, with unofficial reports suggesting possible restoration within days, but no timeline has been announced.
For architects shipping production systems: the export ban signals that frontier AI capabilities are now regulated as national security assets. Models capable of finding code vulnerabilities—regardless of safeguards—face new export friction that affects international teams, cloud deployment decisions, and which frontier tools can be deployed across geographies where non-US employees work.