Anthropic disables Fable 5 and Mythos 5 after a US export control directive

Anthropic has disabled access to two of its models, Fable 5 and Mythos 5, for all customers after receiving an export control directive from the US government. The company says the order arrived at 5:21pm ET on a Friday and that the letter did not spell out the national security concern behind it.

What the directive required

According to the statement quoted by the newsletter 12 Grams of Carbon, the directive ordered the suspension of all access to Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, including foreign national employees of Anthropic. The company calls that an impossible ask: in practice both models had to be switched off for everyone. Other Anthropic models are unaffected.

Anthropic says it understands the government believes it has learned of a way of bypassing, or "jailbreaking", Fable 5. The company reviewed a demonstration of the technique, used to identify a small number of previously known, minor vulnerabilities, and argues those flaws are simple enough that other publicly available models can find them without any bypass. It adds that the capability shown is "widely available from other models (including OpenAI's GPT-5.5)" and is used daily by defenders. Anthropic says it is complying, but disagrees "that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people".

A shadow over the decision

Writing on 12 Grams of Carbon under the name theahura, the author calls himself an AI doomer most days and sets out his worry about powerful optimisers — including corporations, which optimise for what can be measured. He then turns to politics: in his account Anthropic and the administration are "very famously not friends", the company having worked with the Department of Defense before being publicly labelled a supply chain risk, while the Kushners are heavily invested in OpenAI. He notes that Fable is considered a very strong model and that it disappears for consumers just as Anthropic prepares an IPO.

Why the timing and the precedent matter

The author treats the 5:21pm Friday release as suspicious, arguing that bad news is dropped after markets close so the weekend absorbs the shock; he cites a compilation by the research firm The Kobeissi Letter of announcements that landed outside trading hours, from tariff threats to military strikes. The economics matter as much as the politics, he writes: the AI build-out, the debt and the data centres all assume that demand for compute can actually be served, and a government able to cut access unilaterally undermines that assumption. He ends with a Hacker News comment he says matches his own view: the story is not that a lab was punished, but that this may be the start of governments deciding which strong models the public may use.