Two New AI Models Gone Before Most Users Touched Them
Anthropic launched Fable 5 and Mythos 5, then pulled them off the table within days. On Friday night, the company disabled both models entirely – cutting access for all customers worldwide – after receiving a directive from the US Commerce Department that same evening. The directive placed the models under export controls, restricting their use to within the United States only.
Anthropic stated that the only way to guarantee compliance with the government order in the immediate term “is that we must abruptly disable Fable 5 and Mythos 5 for all our customers.”
No other Anthropic models were affected by the shutdown. Claude, and the rest of the company’s existing lineup, remain available.
What the Government Said – and What It Didn’t
According to an Axios report citing an administration official, the Trump administration’s concern centers on a specific jailbreak discovered in Fable 5. The exploit reportedly bypasses the model’s classifier-based safeguards – broad filters designed to block certain categories of prompts. The three domains those filters are meant to protect are cybersecurity, chemistry, and biology. That combination alone explains why this landed on the desk of national security officials rather than, say, a consumer protection agency.
The administration’s stated goal in requesting the pause is to give the “national security apparatus” time to be “hardened” against this type of threat. Axios’ source suggested that process could be finished “in the next few weeks,” though no firm date was attached to that estimate.
Export controls as a mechanism for pausing AI model deployment are not a tool that gets pulled casually. The US Commerce Department’s involvement – rather than a softer request from another agency – signals that the administration treated this as a matter of immediate risk rather than a policy conversation to schedule. Anthropic had no apparent leverage to negotiate a staged rollout or a geographic restriction that stopped short of full shutdown.
What This Means for Users Who Just Got Access
For anyone who signed up specifically to try Fable 5 or Mythos 5, the experience is now a dead end with no timeline attached to reopening. The “few weeks” framing from the Axios source is informal – not a commitment from Anthropic or from any government body. Anthropic’s own statement used the word “abruptly,” which is unusual corporate language and suggests the company had very little runway between receiving the directive and executing the shutdown.
The classifier-based safeguards at the center of this are a standard architectural feature in large language models – essentially rule-based filters sitting in front of the model’s output to catch prompts flagged as dangerous. A jailbreak that gets around broad, category-level classifiers rather than a narrow specific rule is a different kind of problem, because it suggests the bypass works across an entire domain rather than a single edge case.
The shutdown also raises a practical question for enterprise customers who may have already begun integrating Fable 5 or Mythos 5 into internal workflows. Those integrations are now broken, with no migration path offered and no confirmed return date.
Anthropic’s other models continue to operate normally, and the company has not indicated whether Fable 5 and Mythos 5 will return in their current form, be retrained, or be replaced by a hardened version – the government’s next move determines that, not Anthropic’s product roadmap.
