US Lifts Anthropic Model Export Controls After Safeguards Deal
The Commerce Department is removing licence requirements for Anthropic’s Mythos and Fable models after a safeguards agreement, reopening foreign access while leaving jailbreak controls as the unresolved policy test.
Commerce Removes Licence Requirements For Mythos And Fable
The US Commerce Department is lifting export controls on Anthropic’s Mythos and Fable AI models after the company reached a safeguards agreement with the administration, changing a model-access dispute into a compliance test over controlled AI capabilities.
A letter from Commerce Secretary Howard Lutnick to Anthropic cofounder Tom Brown said a licence is no longer required for the export, reexport or in-country transfer of the Mythos or Fable models.
The shift covers both Fable 5 and the more powerful Mythos 5, which had been limited to select companies and government agencies.
The decision follows an earlier order that suspended access for foreign nationals.
The new approval does not remove the policy concern behind that order.
The administration still wants Anthropic to prevent users from bypassing Fable’s safety restrictions to reach capabilities the government treats as sensitive, especially around cybersecurity.
Anthropic Commits To Security Risk Detection
The Commerce letter said Anthropic agreed to proactively detect and address security risks connected to the models.
It also said the company would work with the US government on protocols, standards and releases for Mythos, Fable and future models.
The agreement gives Anthropic a path to restore broader access without proving that every jailbreak can be eliminated.
The company had previously argued that zero jailbreaks were not a realistic standard for a powerful restricted model.
It later emphasized stronger safeguards and a more cooperative process with the administration.
The Wired report names Lutnick and national cyber director Sean Cairncross as officials involved in resolving the dispute.
It also says Anthropic CEO Dario Amodei was replaced in meetings by Brown after officials responded better to Brown’s approach.
Model Access Still Depends On Jailbreak Controls
For enterprise and government buyers, the decision separates access approval from technical proof that restricted capabilities cannot be misused.
The Commerce Department is allowing broader model availability, while Anthropic’s obligations remain tied to detecting and reducing security risks over time.
The policy signal is narrower than a general relaxation of AI export oversight.
That similar restrictions on other AI labs or model families have been removed remain outside the public record.
The public record still lacks the safeguards’ technical benchmarks, audit process, foreign-access monitoring rules or any deadline for proving that jailbreak controls are working.
