When Anthropic is forced to pull the plug on Fable 5 and Mythos 5 on US government orders.

There are Saturday mornings when you expect to quietly scroll through your news feed, coffee in hand, and you stumble onto a statement that reads like the plot of a techno-thriller. The one Anthropic published on the evening of June 12 clearly belongs in that category.

A thought, in passing, for those who upgraded their Anthropic subscription over the past few weeks, precisely to make the most of Fable. The wake-up call is harsh: overnight, the model simply vanished from the interface, without notice. Enough to give this month’s bill a peculiar aftertaste.

Let’s get to the substance. Here, in plain terms, is what the statement says.

What happened#

The US government, invoking national security prerogatives, sent Anthropic an export-control directive requiring it to suspend all access to Fable 5 and Mythos 5 for any foreign national — whether located in the United States or elsewhere, including Anthropic’s own foreign employees. The concrete consequence: to stay compliant, Anthropic has to shut down these two models for all of its customers, without exception. The other models (Opus, Sonnet, Haiku, etc.) are not affected.

The letter arrived on the evening of June 12, without specifying the exact nature of the security concern.

The stated reason: a “jailbreak”#

According to Anthropic, the government believes it has found a method to bypass Fable 5’s safeguards. The company says it has reviewed the demonstration in question: the technique was reportedly used to spot a handful of minor vulnerabilities, already known and relatively simple — so much so that other public models can find them without needing any workaround at all.

At this stage, Anthropic states that it has received only verbal evidence, concerning a narrow flaw: in essence, asking the model to read a specific codebase and fix its defects. The company maintains that this capability is widely available elsewhere — including in OpenAI’s GPT-5.5 — and that it is used every day by the defenders who secure systems.

Anthropic’s line of defense#

The company points to several facts about Fable’s safety:

  • The safeguards are robust — so much so that some users complain they are too broad.
  • Before launch, the model went through thousands of hours of red-teaming with the US government, the UK AISI, third-party organizations, and internal teams.
  • No tester found a universal jailbreak, that is, a method capable of broadly unshackling the model across a whole range of capabilities.
  • Working from the premise that perfect resistance is currently beyond any provider’s reach, Anthropic adopted a defense-in-depth strategy: making flaws either very narrow or very costly to produce, coupled with monitoring that allows attacks to be detected and blocked quickly. This is also what justifies the 30-day data retention policy associated with Fable.

The company’s position#

Anthropic is complying with the directive and removing access to the two models. But the tone is unambiguous: the company disagrees. In its view, recalling a commercial model deployed to hundreds of millions of users over such a narrow flaw would, if the rule were generalized across the industry, amount to freezing the deployment of any new frontier model.

Anthropic acknowledges that the government must be able to block a dangerous deployment — but through a process that is transparent, fair, clear, and grounded in technical facts. Which, the company believes, is not the case here.

The statement closes with an apology to customers, the conviction that this is a misunderstanding, and a promise to restore access as soon as possible. Anthropic also says it will share more details within 24 hours.

To be continued, then. But you have to admit that, for a Saturday morning, we’ve seen more ordinary.

Source: Statement on the US government directive to suspend access to Fable 5 and Mythos 5, Anthropic, June 12, 2026.