Trump admin lifts export curbs on Anthropic's Mythos, Fable

Two weeks after the Commerce Department told Anthropic to pull its two newest models offline, the same department has told the company to switch them back on. According to CNBC, export controls on Claude Fable 5 and Claude Mythos 5 were lifted on Tuesday evening, with customer access returning Wednesday. Commerce Secretary Howard Lutnick said on X that his office had "worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI."

The original order landed on June 12, three days after Anthropic launched Fable 5 as the first publicly available model in its Mythos-class tier. The White House told Anthropic the models posed a national security threat and gave the company 90 minutes to take them down. Anthropic disabled both models entirely rather than try to selectively comply, in part because the directive covered any foreign national anywhere in the world, including the company's own foreign-born staff. Reporting from The Washington Post and others points at a technique for bypassing Fable 5's safety guardrails as the trigger, on the theory that if the filters keeping the model's cybersecurity capability behind an assistant persona can be defeated, the consumer product is effectively an unrestricted cyber tool.

The re-opening did not happen all at once. Access to Mythos 5 was restored earlier for a defined set of roughly 100 US organizations, including government agencies and Fortune 500 companies, for defensive cybersecurity work. TechCrunch reported that carveout on June 26. Tuesday's move removes the broader licensing requirement, and Anthropic has said in the original reporting that Fable 5 is coming back to customers.

The honest caveat is that the retrieved reporting does not spell out what the "appropriate safeguards" Lutnick approved actually are, or what the original bypass technique looked like. Take those specifics as reported, not settled. What is on the record now is the pattern: a US frontier lab launched a model, was told to take it down inside 90 minutes on national security grounds, and got it back through direct negotiation with Commerce. That template is worth watching for OpenAI and Google as much as it is for Anthropic.