The Commerce Department has given Anthropic permission to redeploy its Mythos 5 artificial intelligence model to a limited group of organizations, signaling a partial reversal of restrictions imposed just two weeks earlier on the company's most powerful systems.
Commerce Secretary Howard Lutnick outlined the decision in a letter to Anthropic that was reviewed by NBC News. Lutnick expressed confidence in the security measures Anthropic had implemented to govern access to the system among vetted users. The limited restoration will extend to roughly 100 organizations, including government agencies and private sector companies, according to sources familiar with the matter.
Anthropic confirmed the development in a Friday night post on X, stating that Mythos 5, which the company describes as its strongest cybersecurity model, could now be redeployed to American organizations responsible for defending critical infrastructure. The model had previously been available to a subset of trusted partners through Project Glasswing, which included infrastructure providers and major financial institutions.
The reversal comes after Lutnick invoked export control powers two weeks prior to force Anthropic offline with both Mythos 5 and Fable 5, the company's most recent mass-market model. The administration cited national security concerns, worried that users could potentially circumvent Fable 5's safety guardrails, which were designed to prevent the system from answering questions on sensitive topics in cybersecurity and biological domains.
The export control approach proved sweeping in its impact. Because foreign nationals work at Anthropic and many partner organizations that had accessed the models, the regulations forced a complete shutdown rather than targeted restrictions. The move prompted Anthropic to dispatch senior scientists and engineers to Washington to negotiate with officials at the Commerce Department and the Office of the National Cyber Director.
Those negotiations appear to have yielded partial success. The Mythos 5 restoration signals that government officials believe Anthropic's enhanced security architecture can safely support the model's deployment to critical infrastructure defenders. However, the company is continuing separate discussions about restoring broader public access to Fable 5, the consumer-facing model.
Lutnick's Friday decision arrives just hours after rival OpenAI announced it would phase the release of its new GPT-5.6 family of models in coordination with federal authorities rather than launching them widely at once. OpenAI CEO Sam Altman characterized the staggered approach as unwelcome, noting the company had originally planned a more open launch. In a blog post, OpenAI stated it had shared its trusted partner list with the government ahead of the models' launch and pledged to work with the Trump administration on developing a framework for vetting future models.
The back-and-forth underscores the tension between AI safety concerns and the imperative to maintain access to powerful tools for legitimate purposes. Administration officials had grown increasingly worried about potential circumvention of Fable 5's protections, though some sources said Anthropic's leadership did not fully grasp the severity of those concerns before the restrictions took effect.
OpenAI criticized the current arrangement as unsustainable, writing that ad-hoc government access processes should not become permanent policy because they deny the most advanced tools to users, developers, enterprises, and global partners who need them.
Author Sarah Mitchell: "The partial win for Anthropic buys breathing room, but the larger battle over how Washington controls AI deployment is far from settled."
Comments