The Trump administration has reversed its initial decision to ban Anthropic‘s cybersecurity models, Mythos 5 and Fable 5, by permitting their use to over one hundred specific US government agencies and companies. Commerce Secretary Howard Lutnick confirmed via letter that appropriate safeguards exist to allow trusted partners, including non-American employees at these organisations, to access the technology. This move follows a two-week period where Anthropic withdrew the models after security researchers allegedly bypassed their guardrails. The directive notably excludes Fable 5 from this immediate release, leaving its status uncertain despite it being a version with supposedly stronger protections. Anthropic stated on X that it is restoring access to Mythos 5 for entities defending critical infrastructure while continuing negotiations to make Fable 5 available again.
This policy shift signals a pragmatic approach to maintaining supply chains for essential security infrastructure without waiting for perfect theoretical safeguards. It acknowledges that removing the models entirely created a vacuum in cyberdefence capabilities that the government cannot afford to sustain indefinitely. The selective rollout suggests officials are prioritising operational continuity over a blanket ban on foreign-trained AI models within US borders.
- Access is limited to organisations operating and defending critical infrastructure
- Non-American employees at included entities are now permitted access
- Fable 5 remains excluded from the current directive
