[DEVELOPING]: The US's 'kill switch' export control order on Anthropic pours petrol on AI sovereignty debate

The US government has ordered Anthropic to disable access worldwide to Claude Fable 5 and Claude Mythos 5 over national security concerns.

Anthropic said it complied with the order, issued late on Friday afternoon Eastern Time, but disputed the government’s reasoning. Its other AI models remain available.

Mythos, Anthropic’s most powerful model, had previously been restricted to around 50 organisations because of its ability to identify software vulnerabilities. Participants in the controlled Project Glasswing programme included Amazon, Apple, Google, Microsoft and CrowdStrike.

Fable 5 was released this week as a commercially available version of Mythos, with safeguards designed to block dangerous cybersecurity and biological outputs. Independent benchmarks ranked it as the most capable publicly available model.

Anthropic said the government acted after receiving evidence of a possible jailbreak that allowed Fable 5 to analyse code and identify vulnerabilities. The company argued that the technique was narrow, unconfirmed and already possible with competing public models.

It warned that applying the same standard across the sector could effectively stop frontier AI companies from releasing new models.

The intervention is a significant setback for Anthropic, which has positioned itself as the safety-focused alternative to OpenAI and is reportedly preparing for an IPO. It also exposes the commercial risk of presenting increasingly powerful AI systems as uniquely dangerous, since regulators may take those claims seriously.

Pathfounders will be updating this story today based on the latest events

You can also leave comments on LinkedIn here

This timing could not have been more ironic.

On Thursday morning, a group of European researchers, commentators, and VC investor Judith Dada, released a science fiction-style scenario, “Europe 2031”, part of which is about what would happen if the US pulled the plug on access to frontier AI.

Then yesterday afternoon at 5:21 pm (ET), the US government literally did that. It issued an export control directive suspending all access to Fable 5 and Mythos 5 for any foreign national. The practical effect of this was that Anthropic had to shut down ALL access, globally. Thus, Europeans can no longer access Claude Mythos/Fable.

“Europe 2031” just experienced possibly the fastest time between a work of science fiction being published, and science fact.

Even though Anthropic says it believes the US government's export control order is a “misunderstanding” and that it is working to restore access, the implications are wide-ranging, especially for European countries.

Washington has exercised a ‘kill-switch’ on frontier AI, which other countries have begun to rely on for economic development.

This action just poured petrol on the debate about AI Sovereignty.

• The writers of the Europe 2031 report have today issued a statement saying the US government restrictions on Anthropic are a “wake-up call” and that “Europe must rethink its AI strategy” as a result. 

The move also “sets a deeply worrying precedent” and that “Europe has underestimated how quickly access to frontier AI could become a question of power, leverage and geopolitical bargaining. If anything, today's events make our scenario feel conservative.”

The statement added that Europe now needs a “crisis-level buildout of compute and energy infrastructure, a stronger middle power coalition  with countries such as Japan and South Korea, and faster AI adoption across the economy alongside labour-market reforms that help workers adapt.”

However, they added that “none of this should come at the cost of safety or security. There will be valid reasons to restrict future releases of AI models. But restrictions should be based on transparent, technically grounded assessments and must apply globally - not just to foreign nationals.”