If you wired anything to Fable 5 or Mythos 5 this week, it stopped responding on Friday evening. The cause had nothing to do with your code or your usage. On June 12, 2026, the US government issued an export control directive ordering Anthropic to suspend both models for any foreign national, and the company pulled them for every customer to comply. Every other Anthropic model stayed up.

This is the first time a leading AI company has taken a publicly deployed model offline because of direct federal intervention, so it's worth understanding what it actually changes for the people building on these tools.

What's down, and what still works

The directive bars access by any foreign national, inside or outside the United States, including Anthropic's own foreign-national employees. Anthropic can't reliably tell foreign nationals from domestic users in real time, so the only way to comply was a hard shutoff of both models for everyone, everywhere.

That shutoff isn't limited to Claude.ai. Fable 5 went unavailable on the Claude API (claude-fable-5), and partners pulled it the same day: GitHub Copilot suspended it across all experiences, and Anthropic asked AWS to revoke access on Amazon Bedrock. Mythos 5, the unrestricted version that only vetted partners had, is gone too.

What still works is every other Claude model. Opus 4.8, Sonnet, and Haiku are unaffected on all surfaces. If your workflow was routing to Fable 5, the practical move right now is to point it back at Opus 4.8, which is also where Fable's own safety classifiers were already sending any flagged request.

When does it come back

No firm date yet. In its statement, Anthropic says it believes the order is a misunderstanding and is working to restore access as soon as possible. The only concrete commitment is about information rather than restoration: Anthropic said it would share more details within 24 hours of the Friday 5:21pm ET statement, which points to an update around Saturday evening Eastern time. A restoration date wasn't part of that promise.

If you need to catch the moment it's live again, two places will move first: Anthropic's status page at status.anthropic.com, and the Fable model page, which currently shows the model as unavailable. As of June 13, it's still down across every surface, and nothing public points to a same-week guarantee either way.

Why it got pulled

The government's stated worry, as Anthropic understands it, is a way of jailbreaking Fable 5. The letter itself didn't spell out the concern. By Anthropic's account, the technique it was shown amounted to asking the model to read a specific codebase and fix its software flaws, which surfaced a few previously known, minor vulnerabilities. Anthropic says other publicly available models, including OpenAI's GPT-5.5, find those same flaws without any bypass, and that no universal jailbreak (one that broadly defeats the model's guardrails) has been found. The directive came from the Commerce Department, signed by Secretary Howard Lutnick with input from the Bureau of Industry and Security, according to reporting from NBC News and CNBC.

So the company is complying while contesting the decision in public. Its case: Fable shipped with safeguards red-teamed for thousands of hours with the US government, the UK AI Security Institute, and outside groups, and recalling a model already used by hundreds of millions of people over one narrow vulnerability would, as a general standard, freeze new releases across the whole industry. What Anthropic hasn't published is the government's underlying evidence, because the directive didn't contain it. So far what it has received is verbal.

The part builders should read twice

There's a quieter detail with direct consequences for anyone who sent data through these models. Fable 5 and Mythos 5 launched with a mandatory 30-day data retention policy on all traffic, across Anthropic's own surfaces and third-party ones, and it overrides existing zero-retention agreements. Anthropic says it won't use that data for training and will delete it after 30 days in almost all cases, and that the purpose is detecting jailbreaks that only show up across many requests. Either way, if you ran sensitive prompts through Fable this week believing your usual zero-retention terms applied, they didn't.

That's the real lesson, and it sits one layer below the specific dispute. A capability you depend on can be withdrawn overnight by a decision that has nothing to do with you, your usage, or even the provider's roadmap, and the terms attached to it can differ from the ones you're used to. Treating a hosted model as an assistant whose output you still own and direct has a concrete payoff here: when a third party can switch the assistant off, the value you've built needs to live somewhere that switch can't reach. Keep the parts that are yours, your judgment, your process, your content, portable enough to survive the tool changing under you.

The precedent outlasts the outage

Anthropic has argued publicly that governments should be able to block unsafe deployments through a process that is transparent, fair, clear, and grounded in technical facts, and it says this action didn't clear that bar. Whether Fable 5 returns next week is the question everyone's asking, and it's the smaller one. The thing that sticks is that a live, mass-market AI product can now be switched off by federal order in an afternoon, on evidence the public hasn't seen. That changes how much weight any single hosted model can safely carry in your stack, long after this particular model is back.