In mid-June, the US Department of Commerce moved to block foreign access to Mythos 5 and Fable 5, two frontier AI models from Anthropic, citing national security. The company immediately shut down global access entirely, after being given 90 minutes to comply. The action followed a briefing to Senator Mark Warner in which NSA chief General Joshua Rudd confirmed that Mythos had penetrated nearly all of the agency's classified systems in a matter of hours, The Economist first reported. Technical details remain sealed.

Viewed from Washington, the restrictions crystallise a dual concern: the models' demonstrated offensive cyber capability and the risk of adversarial exploitation. Senator Warner, who disclosed the breach, argued that pre-emptive controls were essential given the technology's power. The Pentagon, which had earlier barred Anthropic from military programmes, saw its stance vindicated – Defence Secretary Hegseth called the exclusion a "correct step" – even as it confirmed using xAI's Grok chatbot in airstrikes on Iran, underscoring a selective embrace of AI tools. In a conflicting signal, President Trump told Axios on Friday that he no longer considered Anthropic a security threat; as of the week's end, however, the access ban remained in place.

European capitals interpreted the episode as a stark warning about technological dependency. According to analysts in Berlin, the shutdown illustrated that access to frontier AI rests on political decisions in Washington. The Frankfurter Allgemeine Zeitung reported that Amazon employees had triggered the intervention by prompting Fable to produce a security-sensitive reply and alerting US authorities, highlighting the opacity of safety enforcement. European leaders meeting AI executives at a G7 summit gained no immediate restoration of access, fuelling calls for accelerated development of homegrown foundation models.

The affair exposes a regulatory vacuum where unilateral US export controls – rather than multilateral frameworks – govern access to transformative technologies. For Anthropic, a company founded on AI safety advocacy, the confrontation with its own government marks a break with its public stance. The breach claims have intensified a global debate about sovereign control of strategic capabilities. Congress is expected to examine the dossier further, and EU officials have signalled they will review partnerships on critical AI infrastructure in the coming quarter.