Statement on the US government directive to suspend access to Fable 5 and Mythos 5

Anthropic Says US Government Directive Will Suspend Access to Fable 5 and Mythos 5

You can read the company’s full statement here: https://www.anthropic.com/news/fable-mythos-access

Anthropic has announced that the US government issued an export control directive requiring the company to suspend all access to Fable 5 and Mythos 5 for any foreign national, whether they are inside or outside the United States. That includes foreign national employees as well. The result is blunt, and a little messy in the way these things usually are, access to both models has to be disabled for all customers, even though access to Anthropic’s other models will continue.

The company says the directive arrived at 5:21 pm ET and did not include specific details about the national security concern. Anthropic believes the government is reacting to a narrow jailbreak technique, one that reportedly involves asking the model to read a specific codebase and fix software flaws. The company says it reviewed the demonstration and found it exposed only a small number of previously known minor vulnerabilities. In their view, other public models can surface similar issues without any bypass at all.

That is where the tension sits. On one side, governments want a way to block unsafe deployments. On the other, companies need rules that are transparent enough to follow without guessing what the standard is today and what it might be tomorrow. Anthropic is saying the current directive misses that balance. They’re complying, but they also argue that treating a narrow jailbreak as grounds to pull a commercial model from broad use would make frontier model releases much harder across the industry.

There’s a practical lesson here for anyone building with AI systems. Access, compliance, and model governance are no longer abstract policy topics. They’re part of the operating plan. One decision from a regulator can ripple through products, customers, and internal workflows in a matter of hours. That’s not theory, that’s Tuesday afternoon for platform teams.

This situation also connects with the broader conversation around AI systems that need the right controls around them, not just raw capability. A useful parallel is Anthropic’s own earlier discussion of access and deployment, which you can see here: https://www.anthropic.com/news/fable-mythos-access

Anthropic says it will share more details within 24 hours and is working to restore access. For now, the bigger question is familiar and unresolved, how do we keep advanced models usable without pretending governance is optional? That answer still feels like it’s being written in real time.