The United States government has barred foreign nationals from accessing Anthropic’s most capable artificial intelligence models, Fable 5 and Mythos 5, prompting the company to disable the systems for all users worldwide. The immediate effect is that researchers, firms and governments outside the US are severed from tools that can autonomously execute complex tasks and, according to US officials, uncover decades-old software vulnerabilities. The restriction, confirmed in an Axios interview with President Donald Trump on 22 June, marks the first time Washington has used export-control logic to deny foreign access to a private-sector AI model at this level of capability.

Anthropic had initially kept Mythos under tight wraps, granting access only to selected American companies after it demonstrated an ability to find security flaws in code that had eluded human reviewers. A lighter variant, Fable, was released more widely until, according to media reports, employees of Amazon elicited a security-relevant response from the system and alerted the administration. Within days, the White House ordered the block. Senator Mark Warner later claimed, as reported by The Economist, that Mythos had breached nearly all classified resources of the National Security Agency within hours. The Pentagon, which had already barred Anthropic from autonomous weapons work, described the expulsion as vindicated. Trump told Axios he no longer views the company as a national security threat, citing its “very quickly” and “responsibly” compliance, but did not rule out invoking the Defense Production Act.

Viewed from Brussels, the episode crystallises a structural vulnerability. Europe’s most advanced foundation model, produced by the French firm Mistral, remains far behind the frontier systems now under US export restrictions. At a G7 summit the same week, leaders met with AI executives but left without a mechanism to restore access. German commentary describes the moment as a wake-up call, accelerating calls for a “CERN for AI” and sovereign European infrastructure. In a parallel but distinct debate, Argentina’s president Javier Milei has proposed granting legal personhood to AI-operated corporations, drawing a rebuttal from historian Yuval Harari, who warned that non-human entities with limited liability could exploit legal vacuums. The exchange, conducted in the Financial Times, underscores the widening gap between technological capability and the legal frameworks available to govern it.

Anthropic and the administration remain in talks to resolve the dispute. The company says it is committed to “protecting critical infrastructure and making sure the U.S. leads in AI.” The next concrete milestone is whether those negotiations produce a revised access framework—or whether Washington extends the controls, forcing a more permanent fragmentation of the AI landscape.