The Trump administration has partially lifted its ban on Anthropic’s advanced cybersecurity AI model, Claude Mythos 5, granting access to over 100 designated U.S. government agencies and critical infrastructure companies. The reversal, communicated in a letter from Commerce Secretary Howard Lutnick on June 27, marks a significant softening of restrictions imposed just two weeks earlier on June 12.
The initial ban ordered Anthropic to withdraw both Mythos 5 and the related Fable 5 model after security researchers demonstrated that the models' guardrails could be bypassed. The ban also prohibited non-American employees, including those at Anthropic, from accessing the models, drawing industry criticism over potential harm to U.S. AI competitiveness.
Under the new directive, Mythos 5 can now be used by a curated list of organizations that operate and defend critical infrastructure. Notably, the authorization extends to non-American employees at those organizations and to Anthropic’s own non-American staff. Lutnick stated that "appropriate safeguards are in place" for trusted partners.
However, Fable 5 remains unavailable for general use. Anthropic is continuing discussions with the government to expand access to Mythos 5 and to restore Fable 5. The company said it is "restoring access for these organizations quickly" and described Mythos 5 as its strongest cybersecurity model, capable of autonomously identifying and patching vulnerabilities.
This calibrated approach signals a shift toward case-by-case AI regulation, balancing national security with operational necessity, and could serve as a template for future governance of powerful AI models.
