Three days. That is all it took for Fable 5 and Mythos 5 — Anthropic's newest and most capable models — to go from launch to a government-ordered shutdown.

The US federal government issued an export control directive citing an alleged jailbreak in the Mythos-class models. Anthropic had no choice but to cut off access globally for all customers. Every other Claude model — including Claude Opus, Sonnet, and Haiku — remains fully operational.

What Happened

Anthropic launched Fable 5 and Mythos 5 on June 10, 2026. Within 72 hours, the US government identified what it described as a jailbreak vulnerability that, in their assessment, created export control risks — meaning the models could potentially be used to generate outputs restricted under US export law.

The directive was immediate. Anthropic pulled both models for every customer worldwide, not just in restricted regions. The company had no public comment at time of publication beyond confirming the action.

What This Means If You Run Claude in Production

If your team uses Claude via the Anthropic API, here is what you need to know right now:

If you were using Fable 5 or Mythos 5 — your API calls to those model strings will fail. You need to pin to a different model immediately. claude-opus-4 or claude-sonnet-4 are the closest alternatives still available.

If you were using any other Claude model — you are not affected. The directive was specific to the Mythos-class models only.

Check your model strings. If your code does not pin a specific model version and defaults to "latest," check what that resolves to right now. Depending on when your dependency was pinned, it may have automatically pointed to Fable 5.

The Bigger Picture

This is the first time a US government export control directive has pulled a commercial AI model mid-deployment. It sets a significant precedent.

For DevOps and platform engineering teams building on top of AI APIs, it is a reminder that model availability is not the same as infrastructure uptime. A model can be technically running and still become inaccessible overnight for legal or regulatory reasons — with zero warning to end users.

The takeaway for production AI systems: treat model versions like external dependencies. Pin them explicitly, monitor for deprecation or access changes, and have a rollback plan that is not dependent on any single model family.

Anthropic has said all other Claude models remain available and unaffected.

Source: The New Stack — original reporting by Matthew Burns, published June 13, 2026.