Are Anthropic's Claude Fable 5 and Mythos 5 gone?

Last Friday evening, Anthropic got a letter from the US government. By the time most people saw the news, Fable 5 and Mythos 5 were already gone, switched off for every user. Both had launched only days earlier, and we'd covered Fable 5 when it did. Now they were pulled worldwide, on a national-security order the company had little choice but to follow.

‍

What happened

The timeline is short. At 5:21pm ET on June 12, an export-control directive arrived from the US government. It told Anthropic to cut off Fable 5 and Mythos 5 for any foreign national, anywhere, including the company's own foreign-national employees. To stay compliant, Anthropic went further than the letter required and disabled both models for every customer worldwide.

Two models, every user.

The government cited national security but, by Anthropic's account, did not spell out the specifics in the letter. Fortune reported the order came from the Commerce Department. One detail matters more than the rest, and Anthropic put it in bold: every other Claude model keeps running. Opus, Sonnet, Haiku, and the earlier tiers are all untouched.

‍

The jailbreak the order points to

The trigger was a way around Fable 5's safeguards. By Anthropic's account, the government learned of a way to jailbreak the model and reach the cybersecurity capabilities those safeguards are meant to keep locked.

Anthropic says it watched a demonstration and wasn't impressed. It saw a handful of minor, already-known vulnerabilities, the kind other public models can find too. By its description, the jailbreak is narrow: ask the model to read a specific codebase and point out software flaws. Anthropic's broader point: no model maker can promise perfect jailbreak resistance today. The same thing works on rival models, it says, including OpenAI's GPT-5.5. And security teams lean on abilities like these every day to defend their own systems.

So the disagreement is about proportion. Anthropic complied, but in the same statement argued that recalling a model used by hundreds of millions, over one narrow jailbreak, would set a bar that stalls every new frontier release.

How people reacted

The directive drew quick criticism. AI-policy and security commentators questioned whether a narrow jailbreak justified pulling a model already in the hands of hundreds of millions, as Fortune reported. Some warned the precedent could slow US AI work. Others pointed out that a company marketing its models as near-weapons invites exactly this kind of attention.

The timing stands out too. This comes barely a week after Anthropic filed to go public at a roughly $965 billion valuation, and a regulator switching off a flagship product overnight is the kind of risk investors now have to price in.

What it means if you were building on them

For most teams, the direct hit is small. Fable 5 and Mythos 5 launched days ago, so few products or workflows were leaning on them yet, and everything else from Anthropic still works.

But the bigger lesson is simple. A model you build on can disappear for reasons that have nothing to do with you or your usage, on a few hours' notice.

If you run marketing or product features on a frontier model, a few things are worth checking this week:

• Map your single points of failure. Know which live features or workflows call one specific model, and what breaks if that model goes dark tomorrow.
• Keep a fallback wired. A second model you can switch to, even at lower quality, beats a scramble during an outage.
• Check the data terms. Anthropic held 30 days of Fable customer data as a safeguard, which mattered for anyone who ran a pilot on it.
• Separate the model from the vendor. Other Claude models kept working here, so "build on Anthropic" and "build on one Anthropic model" are different bets.

We keep the AI workflows we run model-portable for exactly this reason. Any single model can go away, and the work shouldn't go with it.

What we're watching

A handful of open questions will decide how much this matters.

‍

The two models stay dark until the directive lifts, and nobody has put a date on that. Worth watching whether the order stays narrow or spreads to other models and other labs, and whether it cools the mood around that IPO. We'll keep tracking it from the desk. If a model vanishing overnight would break something you've built, pressure-test your stack with us before it does.

More once the directive moves either way.