US forces Anthropic's Fable 5 offline, allies call it 'kill-switch'

Key Takeaways

• The US Commerce Department forced Anthropic to disable Fable 5 and Mythos 5 globally on June 12, citing national security after a reported jailbreak.
• European and Canadian leaders are calling the shutdown a wake-up call, with several demanding investment in sovereign AI infrastructure.
• This marks the first time the US government has used export controls to force the recall of a commercial frontier AI model.

Three days after Anthropic released Fable 5 to the public, the US government pulled it offline worldwide. The Commerce Department issued an emergency directive on June 12 ordering the company to disable both Fable 5 and its predecessor Mythos 5, citing national security concerns after reports the model had been jailbroken. It is the first time American export controls have been used to force the recall of a commercial frontier AI model.

The shutdown did not just affect American users. British hospitals piloting the technology lost access. Government projects across Europe went dark. Canadian research initiatives halted. The global nature of the ban has transformed what could have been a domestic regulatory story into an international incident, with allied leaders now openly questioning whether reliance on US-based AI providers constitutes a strategic vulnerability.

How did Fable 5 and Mythos 5 get here?

Anthropic launched Mythos in April as a cybersecurity-focused model, restricting access to a handful of organizations under Project Glasswing. The company claimed the model was too capable to release broadly, positioning it as a tool for finding vulnerabilities in legacy codebases. The marketing leaned heavily on fear: Anthropic warned that similarly powerful models would be available within 18 months, so organizations needed to prepare their defenses now.

In early June, Anthropic expanded Mythos access to 150 organizations globally. Days later, it dropped Fable 5, a model with Mythos-grade capabilities but additional safeguards designed to prevent misuse for offensive cybersecurity tasks. The safeguards did not last long. The US government claimed Fable 5 had been jailbroken and moved to shut it down entirely.

By June 12, both models were offline and inaccessible. Heads of US tech firms including Nvidia and Adobe have reportedly been in talks with the Trump administration to reinstate access, arguing the bans actually hamper defensive cybersecurity efforts. So far, the administration has not budged.

European leaders frame this as digital sovereignty

The response from European politicians has been blunt. French MEP Christophe Grudler called the shutdown proof of a long-standing concern: "The United States is once again demonstrating what we Liberals and Democrats have warned about so many times since Trump entered into office; that the US holds a real 'kill-switch' over essential technologies and that they are more than willing to use it."

Dutch MEP Bart Groothuis was equally direct. "These restrictions are a clear example of the current American 'nobody but us' mentality," he said. "Once again: this shows that Europe needs its own LLMs and open weight models or face digital colonization."

The framing matters. Politicians are not treating this as a one-off enforcement action but as evidence that frontier AI access is now critical infrastructure, on par with energy or telecommunications. That reframing carries significant policy implications: it suggests governments should not be comfortable depending on foreign commercial providers for AI capabilities, regardless of how open the market appears in normal times.

Canada takes a softer tone but the same lesson

Canadian Prime Minister Mark Carney avoided directly criticizing the US. "Nobody has done anything wrong in the situation," he said. But his conclusion pointed in the same direction as his European counterparts: "We will have done something wrong if we just accept this, don't take the lesson, don't build out and diversify."

The diplomatic language differs, but the policy prescription is identical. Countries that assumed market access would always be available are now reconsidering that assumption. Even close allies are talking about building domestic AI capabilities specifically to reduce dependence on American firms.

The UK's truncated pilot programs

Former UK Armed Forces Minister Al Carns put the impact in concrete terms: British researchers were studying the models, British companies were testing them, and British hospitals were piloting them. All of that stopped when a foreign government flipped a switch.

Carns argued this was exactly why the UK needed to develop its own frontier AI tools rather than depending on American providers. The UK has significant AI research expertise, he noted, and should be using it to ensure sovereign access to capable technologies.

What makes this export control different

Export controls on technology are not new. The US has long restricted sales of advanced semiconductors and manufacturing equipment to certain countries. But those controls typically prevent future sales or transfers. This is different: the government ordered an American company to disable software that was already running in allied countries.

Cloud-native AI does not work like physical goods. You cannot just stop shipments at the border. When the government wanted Fable 5 offline, it had to go offline everywhere. That global reach is precisely what worries allied governments. If the US can disable AI infrastructure worldwide with a single directive, the risk calculus for depending on American providers changes significantly.

What happens next

In the short term, programmers on Reddit and Claude subreddits are hoping access will be restored. The tech industry lobbying effort continues. But the broader damage may already be done. The assumption that market access to frontier AI was reliable has been shattered.

Longer term, expect increased investment in non-American AI development. The EU has been talking about AI sovereignty for years, but those discussions were often abstract. This incident gives them a concrete example to point to. Open-weight models, which cannot be disabled remotely, will likely see increased interest from organizations that cannot afford access to be revoked overnight.

Frequently Asked Questions

Why did the US government shut down Fable 5?

The Commerce Department claimed Fable 5 had been jailbroken and posed national security risks. The emergency directive on June 12 forced Anthropic to disable both Fable 5 and Mythos 5 globally.

Is this the first time the US has used export controls on an AI model?

Yes. This is the first instance of the US government using export controls to force the recall of a commercial frontier AI model that was already deployed.

Will Fable 5 access be restored?

Unknown. Tech executives from Nvidia and Adobe are reportedly lobbying the Trump administration to reinstate access, but no timeline has been announced.

What are European countries doing in response?

Several European leaders are calling for investment in sovereign AI infrastructure and open-weight models to reduce dependence on US-based AI providers.

What is Project Glasswing?

Project Glasswing was Anthropic's initial restricted access program for Mythos, providing the model to select organizations to improve code security before the wider release.

Source: Latest from Tom's Hardware