Anthropic’s most advanced artificial intelligence systems are set to return to public access this week after Washington lifted a short-lived export ban imposed over fears the technology could be weaponised, Reuters reported on Tuesday.
The US Commerce Department had ordered the company on June 12 to immediately block foreign access to its Fable 5 and Mythos 5 models. Anthropic, unable to verify users’ nationalities in real time, pulled both offline entirely. The trigger was a discovery by Amazon researchers that Fable 5’s safeguards could be bypassed to find software vulnerabilities and, in one case, generate code that exploited one.
Anthropic said it has since added a new safety measure that catches such requests and silently redirects them to a weaker model, Opus 4.8. As a result, Fable 5 will be available to the general public from Wednesday. Mythos 5, designed specifically to detect cybersecurity weaknesses, had already been cleared last week for a small set of trusted US organisations and will now expand to more domestic and international partners under the Glasswing programme.
The company acknowledged the fix is not seamless, warning that some harmless queries will be blocked and that making any AI completely immune to jailbreaks is “probably impossible.” No universal jailbreak has been discovered yet, but external red-team testers continue probing the model.
Commerce Secretary Howard Lutnick stated in a letter seen by Reuters that the department could reimpose restrictions at any time if Anthropic fails to honour its commitments. The company has agreed to report malicious activity and work with the government on protocols for future releases.
The episode is part of a broader White House push to review frontier AI models before they reach wider audiences. President Donald Trump recently signed an executive order encouraging developers to submit powerful new systems for up to 30 days of government assessment. Rival OpenAI confirmed last week that it had delayed a full public release of its GPT-5.6 model at Washington’s request.
According to Reuters, Anthropic is now deepening its ties with the US government, granting designated agencies early access to both models, and working with Amazon, Microsoft, Google, and other Glasswing partners to create shared standards for evaluating and fixing jailbreaks.
The Pentagon earlier this year labelled the company a “supply-chain risk” after it refused to permit its AI to be used for mass domestic surveillance or fully autonomous weapons. Both Anthropic and OpenAI have confidentially filed for initial public offerings.