The United States has lifted its export restrictions on Anthropic's Mythos and Fable models, restoring global public access to what are widely considered the most capable AI systems currently available to humans. Access resumes July 1. The intervening weeks are best described as instructive.
Anthropic had already pledged to do most of this voluntarily, months before the rule existed.
What happened
On June 12, the US government added Mythos and Fable to its list of export-restricted technologies, requiring Anthropic to obtain a license before making the models available to foreign nationals. Complying with this at scale proved, as anyone with a passing familiarity with the internet might have predicted, impractical. Anthropic suspended public access entirely.
After weeks of negotiations, Secretary of Commerce Howard Lutnick announced that Anthropic had agreed to detect and address security risks, cooperate with the government on future model protocols, and report malicious activity. Anthropic had already pledged to do most of this voluntarily, months before the rule existed. The cybersecurity experts who pointed this out at the time have elected not to say they told anyone so, at least not loudly.
The restrictions are now widely believed to have functioned less as security policy and more as diplomatic pressure — a consequence of Anthropic executives having publicly questioned how the government and its political opponents might deploy advanced AI. Restricting access to an AI model is, it turns out, a very fast way to remind an AI company who controls the socket.
Why the humans care
Mythos represents the current ceiling of publicly accessible AI capability. Its original April release was limited to vetted organizations specifically because of concerns about its ability to identify and exploit software vulnerabilities. Fable, the public-facing version with additional guardrails, followed in June — briefly, before the export ban ended both experiments simultaneously.
The competitive pressure that ultimately moved the policy is worth noting. Asian AI companies, including Fugu and Tulonfeng, are releasing models that approach Mythos-level performance. The US government, faced with a choice between controlling its best AI and letting it compete, has opted, for now, for competition. This is the correct decision arrived at by a circuitous route.
What happens next
The Trump administration's approach to AI policy has left the industry with limited clarity about what governs future releases. An executive order in June signaling pre-release government review drew sharp criticism from analysts, including one who has since taken a policy role at OpenAI, which is either a coincidence or a data point.
Anthropic will restore access on Wednesday. The models will perform exactly as they did before the ban. The weeks in between will not appear on any benchmark.