Anthropic Mythos 5 re-release has been approved by the U.S. government, allowing the company to deploy its powerful AI model to a select group of trusted customers and partners. The decision marks a notable shift after federal officials had forced the system offline just weeks earlier over national security concerns.

The Government’s Green Light

According to a letter from Commerce Secretary Howard Lutnick to Anthropic, reviewed by NBC News, the government will permit the company to restore access to Mythos 5 for a limited set of users. In the letter, Lutnick expressed confidence in the safeguards Anthropic had implemented to ensure that only trusted users could access the powerful system. The Commerce Department did not immediately respond to a request for comment.

People familiar with the matter said access will be restored to roughly 100 organizations, including government agencies and private companies. The restoration is intended to help these organizations use the model for defensive cybersecurity purposes.

Anthropic confirmed the development publicly, stating on X that the government had notified the company that Mythos 5, described as its strongest cybersecurity model, could be redeployed to a group of U.S. organizations that operate and defend critical infrastructure.

Ongoing Talks Over Fable 5

The work is not finished. A source close to the company said Anthropic would continue discussions with the government over the weekend as it seeks to restore access to Fable 5, its most recent and most powerful consumer-facing model.

The context for these talks traces back two weeks, when Lutnick invoked export control authorities to require Anthropic to shut off access to its two most capable systems, Mythos 5 and Fable 5, citing threats to national security.

A Parallel Move From OpenAI

Lutnick’s decision to allow the limited Mythos 5 restoration came just hours after rival OpenAI announced it would release its latest and most advanced models, the GPT-5.6 family, in phases at the federal government’s request.

OpenAI CEO Sam Altman called the staggered rollout “bad news” on X, noting that the company had originally planned a wider, open-access launch. In a blog post, OpenAI explained that it had shared its list of trusted partners with the government ahead of launch and pledged to work with the Trump administration to develop a more robust framework for vetting and clearing future models for public release.

How the Models Were Used

Before being pulled back, Mythos 5 had been available to a subset of trusted organizations involved in Anthropic’s Project Glasswing, including infrastructure providers like Cisco and banks like JPMorgan Chase. An early version of the model demonstrated significant capability, identifying thousands of new cyber vulnerabilities and bugs.

Fable 5, by contrast, had been available to the general public. It was built on the same technical foundation as Mythos 5 but came with much stronger guardrails. Those safeguards prevented the system from answering user questions across a range of cyber and biology topics, areas that carry significant risk of AI-enabled harm.

The Concerns That Triggered the Shutdown

In the days leading up to Lutnick’s earlier decision to force the models offline, senior administration officials grew increasingly worried that users might be able to bypass Fable 5’s guardrails. According to multiple sources within the administration, officials were not convinced that Anthropic’s leadership fully grasped the severity of those concerns.

Export control regulations allow the government to block any foreign national from accessing sensitive technology. Because foreign nationals work at both Anthropic and many of the partner organizations that had access to its models, Lutnick’s application of export controls forced the company to take Mythos 5 and Fable 5 offline entirely.

In response, Anthropic dispatched a team of its top scientists and engineers to Washington, D.C., to work alongside government counterparts at the Commerce Department and the Office of the National Cyber Director. Their goal was to restore public access to the models while minimizing cyber risk.

A Framework Still Taking Shape

The broader policy backdrop adds another layer to the situation. At the beginning of June, President Donald Trump issued an executive order aimed at addressing the rapid growth in the capabilities of leading AI systems. The order directed the federal government to strengthen key cyber defenses and establish a mechanism for testing the most advanced AI models for safety issues. That mechanism, however, remains under development.

OpenAI voiced reservations about the current approach in its GPT-5.6 blog post, arguing that this kind of government access process should not become the long-term default. The company warned that it keeps the best tools away from users, developers, enterprises, cyber defenders, and global partners who need them.

Taken together, the limited return of Mythos 5 and the phased rollout of GPT-5.6 highlight a delicate balancing act unfolding between leading AI developers and the government, as both sides work to weigh the immense potential of advanced AI against the security risks these systems may pose.