Fable 5 and Mythos 5, two of Anthropic's newest AI models, went offline after US officials raised concerns about a possible safety bypass, exposing a regulatory gap no one has answered yet.
Anthropic took two of its newest AI models, Fable 5 and Mythos 5, offline this week after the US government raised concerns about a possible jailbreak, a demonstrated way to bypass the safety limits built into a model. The company says the only compliance option it was given was to pull the plug, not restrict certain users and not add guardrails, but to remove the models from service entirely.
The trigger was a federal export-controls action, according to a Monday report from CNET's Jon Reed. Export controls are a Cold-War-era legal tool designed to keep physical and dual-use technology from reaching adversaries. They are built around things that move through borders: chips, missile components, encryption hardware. In the modern reading, they also reach US-origin software accessed by foreign nationals anywhere in the world, which is why a consumer-facing AI model can trip the rule even when no physical object ever crosses a customs line. The national-security instrument the government used here does not have a surgical setting for an artifact that can be copied by anyone with a URL.
That is the structural problem at the heart of this story, and it is older than any single company. The legal mechanism the government used to justify a model shutdown was designed before AI software existed, and the question of what process should govern a frontier model release with national-security concerns does not have an answer yet. CNET characterizes the episode as a renewed Anthropic-versus-Washington conflict, a pattern claim that the available same-day reporting does not anchor to a specific prior incident.
A second note of caution on the facts. The "jailbreak" framing is CNET's paraphrase of an unverified federal claim, not a confirmed exploit, and Anthropic disputes the government's read of the demonstration. Both things are true at once: the company disagreed with the substance and complied with the action. The CNET headline groups the pulled offerings as "Claude Fable and Mythos AI Models," but the visible reporting substantiates the Fable 5 pull in detail. Mythos's specific role in the shutdown, and any prior access restrictions on it, remain to be confirmed by Anthropic's own statement or additional reporting.
Anthropic's own framing, as quoted in the CNET piece, is the most useful sentence in the story. The company is asking for a statutory process that is "transparent, fair, clear, and grounded in technical facts." That framing concedes the government's right to act while contesting the mechanism. The constructive read is that we have a regulatory gap to fill, not a regulatory fight to win.
Same-day timing means downstream facts are likely to land in the next 24 to 72 hours. Anthropic's own statement, the federal source for the jailbreak claim, and the technical specifications of Fable 5 and Mythos 5 are all pieces of the picture that have not yet been published. If a single contested jailbreak demonstration can trigger a full shutdown of a frontier model, the chilling effect on any lab shipping one is a substantive policy question rather than industry boilerplate. It is the question this story will be answering in the weeks ahead, once more of the record is on the page.