How Anthropic’s Mythos 5 Became A US-Canada-EU Diplomatic Crisis

Senior Anthropic staff were in Washington on 15 June meeting White House officials in an attempt to resolve a dispute that, within the space of three days, had escalated from an internal security concern to a transatlantic diplomatic incident.

The trigger was the Trump administration’s decision to impose export controls on Anthropic’s two most advanced models, Fable 5 and Mythos 5, blocking foreign nationals from accessing them. Anthropic responded by disabling both models globally for all users while negotiations continue.

Canada’s Prime Minister Mark Carney has compared the situation to the systemic risks that caused the 2008 financial crisis. The EU Commission has warned that any contingency measures “should not be discriminatory against partners.” And the company at the centre of it all is in a meeting room in Washington, trying to get its own models switched back on.

What Fable 5 And Mythos 5 Are

Anthropic launched Fable 5 on 9 June, describing it as the first publicly available Mythos-class model.

Fable 5 and Mythos 5 share the same underlying ten-trillion parameter architecture, with Fable 5 carrying additional safety classifiers. Both were previously restricted to government agencies and a small number of organisations in Anthropic’s Project Glasswing programme before the 9 June release made Fable 5 widely available.

The models are state-of-the-art on virtually all AI capability benchmarks, with a one-million token context window and, critically, advanced cybersecurity capabilities that include the ability to construct full attack chains and identify vulnerabilities. This last point forms the basis of the current dispute.

Amazon researchers were able to use prompts to get Fable 5 to provide information that could facilitate cyberattacks, information that was supposed to be blocked by the model’s safety protections. Amazon CEO Andy Jassy briefed Treasury Secretary Scott Bessent and other officials on those findings – the export control directive followed within days.

How A Security Concern Became A Geopolitical Dispute

The escalation’s rapid pace is the key detail. On 12 June, the Trump administration issued a directive blocking foreign nationals from accessing Fable 5 and Mythos 5. By 12-13 June, Anthropic had disabled both models globally, not only for foreign users, while it worked out how to comply. Anthropic was given no specific national security details about the concern that prompted the directive.

Without knowing what triggered the directive, a company can’t easily fix or appeal it. When the alternative is accidentally breaking export control laws, shutting down access globally is simply the safest move. Consequently, global businesses, researchers and developers abruptly lost access to two frontier models.

Canada sent a clear, sharp message in response. Speaking at the G7, Prime Minister Mark Carney said: “The situation we’re in collectively right now with Mythos and Fable is something that can happen with over-reliance on certain models… we have similar things in terms of model risk,” explicitly drawing a parallel to the systemic bank linkages that amplified the 2008 financial crisis.

The EU Commission was more measured but equally clear, warning through spokesperson Thomas Regnier that measures “should not be discriminatory against partners.”

The Bigger Question For Anyone Who Relies On Frontier AI

The Mythos 5 dispute has transformed a longstanding theoretical concern into a tangible reality.

Access to frontier AI models can be switched off, instantly, based on a single company raising concerns with government officials. The export controls apply to foreign nationals regardless of where they are located, meaning a developer in Paris, a researcher in Singapore or a startup team in London using these models through an API could find access removed without warning and without explanation.

This isn’t entirely new territory. Export controls have long applied to chips and hardware, and US cloud infrastructure has always carried some degree of geopolitical exposure. What’s different now is that the models themselves, the software layer that operates above the hardware, are being treated as controlled technology. The precedent being set in Washington this week will have implications beyond Anthropic’s current dispute with the administration.

For any business that has built workflows, products or research pipelines around a single frontier model, the question the Mythos 5 situation raises isn’t really about Anthropic specifically. It’s about the assumption that access to these models is a stable, reliable utility, and whether that assumption has ever been as solid as it seemed.