Anthropic Fable 5 on Track to Return Within Days, Sources Say

Fifteen days after the U.S. government ordered Anthropic to shut down its most powerful AI models, sources told Axios that restrictions on Fable 5 could lift as soon as this coming week. A second source noted that conversations between Anthropic and the administration are expected to continue over the weekend, though the Pentagon and National Security Agency still need to give Fable 5 the green light, leaving the outcome unpredictable.

The clearest signal of progress came Friday, when the Commerce Department allowed Anthropic to restore access to Mythos 5, described as the company's strongest cybersecurity model, to certain trusted U.S. organizations, partially reversing the earlier suspension. Fable 5 and Mythos 5 share the same underlying model architecture; the distinction is that Fable 5 is designed for widespread public availability while Mythos has its safeguards lifted for specific applications.

The ban itself stems from an export control order the government issued on June 12, citing a jailbreak vulnerability in Fable 5. Anthropic reportedly characterized the vulnerability as narrow, one that would unlock Mythos's cybersecurity capabilities in only one specific instance, not a universal bypass, and said it believed the government's concern was a misunderstanding.

The honest caveat in the Axios reporting is that unnamed sources can be wrong, and the administration's two key security agencies have not yet signed off. Neither Anthropic nor the White House immediately responded to requests for comment. What the reporting does not give you is any detail about what assurances or technical changes, if any, Anthropic has offered to secure approval, or whether Fable 5 will return with the same international availability it had before June 12.

If Fable 5 does return this week, Anthropic closes out one of the stranger episodes in commercial AI history: a major model pulled not by the company's choice but by government directive, and restored on a timeline largely outside the company's control. For the broader industry, the episode has already demonstrated that export controls can reach individual AI models, a form of regulatory authority the sector had not previously tested in public.