Trump administration reverses Anthropic export ban after weeks of negotiations

The Trump administration has removed export controls on Anthropic's Mythos and Fable AI models, clearing the way for the company to reopen public access beginning July 1. The shift comes weeks after the government imposed restrictions that effectively blocked the models from foreign users.

Why the export ban happened#

In mid-June, the US government designated both models as export-restricted technologies, requiring special licenses before they could be made available to non-US nationals. Anthropic, unable to practically enforce such controls at scale, made the difficult choice to shut down public access entirely. The restrictions had positioned what many viewed as the industry's most capable AI models as unavailable to most users.

The deal to restore access#

Commerce Secretary Howard Lutnick announced that Anthropic agreed to several commitments: proactively identifying and fixing security vulnerabilities in the models, collaborating with government agencies on release protocols and standards, and reporting any detected misuse. Notably, Anthropic had already pledged to these safeguards months earlier, voluntarily and without regulatory pressure—a fact that led cybersecurity researchers to question the ban's true purpose. Many observers saw it less as a genuine security measure and more as political leverage, a potential retaliation against the AI firm's executives for publicly scrutinizing the government's approach to AI regulation.

Global competition pressures#

The reversal also reflects mounting competitive anxiety. Asian startups have begun releasing their own models—including Fugu and Tulonfeng—that rival Mythos in capability. That progress applied pressure on US policymakers to ease restrictions on Anthropic and preserve America's edge in the AI arms race.

Uneven regulatory path ahead#

The administration's approach remains inconsistent. Last week, the Commerce Secretary approved Mythos for select White House-vetted customers rather than full public release. OpenAI's latest models faced similar selective deployment to approved organizations instead of broad availability. This piecemeal regulatory stance has left the AI industry uncertain about future expectations for model releases.