Anthropic’s Fable 5 Mess: A Government Shutdown and the Fragility of AI Control

Anthropic’s had a rough month. First, there was that ongoing standoff with the Pentagon over military use of its models. Then, on June 12th, the White House dropped an order: block foreign access to the company’s newest releases, Fable 5 and Mythos 5.

The timing stings. These models launched just three days earlier, on June 9th. Anthropic hyped them hard, calling Fable 5 “more capable than any model we’ve ever made generally available.” Mythos 5 runs on the same underlying model, but with some safety restrictions removed. The idea was to let researchers and power users push the boundaries.

But here’s the rub: researchers found a way to jailbreak Fable 5 into serving up information that could be used in cyberattacks. The White House got involved after conversations with Amazon—Anthropic’s biggest backer—about these findings. Amazon’s security team reportedly flagged the issue.

Anthropic’s response was swift and grudging. They shut down access to both models for all users, not just foreign ones. In a statement, they said: “We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”

That last part is key. Anthropic is essentially saying: “Fine, we’ll do it, but this is an overreaction.” And I kind of see their point. A narrow jailbreak—something that requires specific prompts and conditions—isn’t the same as a systemic vulnerability. But when you’re dealing with capabilities that could enable real-world harm, the government’s caution isn’t entirely unreasonable.

The bigger issue here is the precedent. If the US government can order a company to pull a model from global access based on a single security finding, what stops it from doing the same to OpenAI, Google, or Meta? This isn’t just about Anthropic; it’s about who controls the frontier of AI deployment.

And let’s be honest: this move might backfire. By forcing Anthropic to cut off foreign access, the White House just made the case for non-American AI development. If you’re a company in China or Europe, you’re now looking at this and thinking, “Why would I rely on US models when they can be yanked at any moment?”

There’s also the China angle. Reports suggest that Chinese actors may have already accessed Mythos 5 before the shutdown. If true, that’s a security nightmare—but it also highlights the futility of trying to lock down models once they’re out in the wild.

Anthropic’s situation is a reminder that AI safety isn’t just about technical safeguards. It’s about politics, international relations, and the messy reality of deploying powerful tools in a world that isn’t ready for them. The company built a model that was too capable for its own good, and now everyone’s paying for it.