The US government just pulled Anthropic's most powerful AI. This has never happened before

Three days. That's how long Claude Fable 5 lasted before the US government pulled the plug.

On Monday, Anthropic released what it called the most capable public AI model ever made. By Thursday evening, the Commerce Department had ordered the company to shut it down for every single customer. No phased rollback, no grace period. Just off.

This has never happened before. A government has never forcibly recalled a live commercial AI model.

What happened

At 5:21pm ET on Thursday, Anthropic received an export control directive from Commerce Secretary Howard Lutnick's office. The order required Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States. That includes Anthropic's own foreign national employees.

Because the company can't reliably distinguish between domestic and international users in real time, it did the only thing it could: disable both models for everyone.

The trigger, according to Axios, was a rival company claiming it had found a way to jailbreak Mythos 5. The Trump administration, already nervous about the model's cybersecurity capabilities, took that claim and ran with it.

Anthropic is pushing back

Anthropic complied with the directive, but its public statement reads less like corporate compliance and more like a company that thinks it's been wronged.

The company says it reviewed the jailbreak in question and found it was narrow, not universal. Essentially, someone got the model to read a codebase and flag some known, minor software vulnerabilities. Anthropic points out that other publicly available models can do the same thing without any jailbreak at all, including OpenAI's GPT-5.5.

Before launch, Anthropic ran over 1,000 hours of red-teaming with the US and UK governments, private organisations and internal teams. No tester found a universal jailbreak. The company acknowledges that no model has perfect jailbreak resistance, which is exactly why it built a defence-in-depth strategy with aggressive monitoring and mandatory 30-day data retention.

In other words, Anthropic is saying: we told you the guardrails weren't perfect, we explained our mitigation plan, you tested it yourselves, and now you're pulling it because someone else found a narrow edge case that other models also have.

Why this matters

Forget Anthropic for a second. Think about what this means for the industry.

If a narrow, non-universal jailbreak is enough to trigger a full model recall, then no frontier model is safe from the same treatment. Anthropic said as much in its statement: applying this standard across the industry would "essentially halt all new model deployments for all frontier model providers."

That's not hyperbole. Every major AI model has known jailbreak vulnerabilities. If the bar for a government recall is "someone found a way to make it do something it shouldn't in a specific scenario," then GPT-5.5, Gemini and every other frontier model is one bad-faith report away from the same fate.

For developers and businesses that built on Fable 5 this week, the message is brutal. Your production model can vanish overnight because of a political decision with no prior notice, no technical review process, and no clear standard for what triggers it.

What happens next

Anthropic says it believes this is a misunderstanding and is working to restore access. It also made clear it thinks the government should have this power, just through "a statutory process that is transparent, fair, clear, and grounded in technical facts." A polite way of saying: not like this.

The next 24 hours matter. Anthropic promised more technical details about the jailbreak. If the company can demonstrate convincingly that the capability in question is already available in competing models, the Commerce Department will be in an awkward position. You can't credibly argue Mythos 5 is a unique national security risk if GPT-5.5 does the same thing and is still live.

This story is far from over. But the precedent is already set. The US government can and will pull a commercial AI model from the market on short notice. Every AI company just learned that the hard way.