Washington pulls the plug on Anthropic's frontier models
On June 12, Anthropic said the US government invoked national-security export-control authorities to suspend all access to Fable 5 and Mythos 5 - not just for overseas users, but for any foreign national anywhere, including Anthropic's own foreign-national staff. In practice, that forced the company to switch both models off for everyone to stay compliant. Crucially, the rest of the Claude lineup keeps running.
What the government is actually worried about
The directive landed at 5:21pm ET and, per Anthropic, didn't spell out the specific concern. The company's read is that officials believe someone found a way to "jailbreak" Fable 5 - bypass its safety guardrails. Anthropic says it reviewed a demonstration of the technique and found it surfaced only a handful of previously known, minor software flaws, the kind that other publicly available models (including OpenAI's GPT-5.5) can dig up without any bypass at all.
Anthropic's pushback, in plain terms
Anthropic is complying, but it clearly disagrees. A few threads from its statement:
- The company argues that the cited narrow jailbreak - essentially asking the model to read a codebase and patch its flaws - describes a routine task defenders rely on every day, not a smoking gun.
- It leans on a defense-in-depth approach: guardrails tuned so conservatively that users complained, thousands of hours of pre-launch red-teaming with the US and UK governments, and a 30-day data-retention policy meant to catch and shut down attacks fast.
- No tester, it says, has yet found a universal jailbreak that broadly unlocks dangerous capabilities - and it maintains that none of the disclosed issues offered meaningful Mythos-specific uplift.
Why this is bigger than one company
Here's the part that should make every frontier lab pay attention: Anthropic warns that if a single narrow jailbreak became the standard for recalling a model already serving hundreds of millions of people, it would effectively freeze new model releases across the entire industry.
The company also threads a needle on its own policy. It has publicly argued that governments should be able to block unsafe deployments - but only through a process that's transparent, fair, and grounded in technical facts. Its message here is that this particular action didn't clear that bar.
What customers should do now
- Both Fable 5 and Mythos 5 are off until further notice; every other Claude model is unaffected.
- Anthropic framed this as a likely misunderstanding, said it is working to restore access, and promised more detail within 24 hours.
- If your workflows lean on Fable-class capability, the practical fallback is Opus 4.8 - which already fields many sensitive queries through Fable's safeguard routing anyway.
Is this a one-off dispute, or the first real test of government "kill-switch" authority over a commercial AI model? Either way, it's a live preview of how messy that power can get.