Anthropic’s most powerful AI model is back online at least for a carefully vetted group of users. The U.S. government has granted the company permission to redeploy Mythos 5, its most advanced cybersecurity-focused AI, to approximately 100 organizations across both the public and private sectors. The move, authorized through a letter from Commerce Secretary Howard Lutnick, signals a renewed level of institutional confidence in Anthropic’s approach to responsible AI deployment.
The clearance arrives after a tense two-week period during which the federal government had suspended access entirely amid concerns about the technology falling into the wrong hands.
Why the government pulled access in the first place
The suspension was not a minor procedural pause. Secretary Lutnick invoked export control authorities to halt access to both Mythos 5 and a separate model, Fable 5, after officials raised concerns that users could potentially find ways around the safety guardrails built into the systems.
Adding complexity to the situation, the government expressed unease about the presence of foreign nationals within Anthropic and its network of partner organizations. Those concerns prompted strict controls over what officials characterized as sensitive technology with significant national security implications. The action underscored just how seriously federal regulators are beginning to treat frontier AI models not merely as software products, but as tools with real geopolitical weight.
Who gets access and why it matters
The organizations now cleared to use Mythos 5 are primarily those involved in defending the country’s critical infrastructure. Anthropic confirmed the development in a post on X, noting the government’s authorization to redeploy the model to entities that operate and protect essential systems nationwide.
Before the suspension, Mythos 5 had already been in use by a select group of trusted participants in Anthropic’s Project Glasswing, a program that includes major infrastructure providers such as Cisco and financial institutions including JPMorgan Chase. An earlier version of the model had demonstrated meaningful real world utility by identifying thousands of previously unknown cybersecurity vulnerabilities and software bugs the kind of capability that defenders of critical systems are eager to put back to work.
How Mythos 5 and Fable 5 differ
Not all of Anthropic’s models faced the same path forward. While Mythos 5 has now been cleared for a targeted group of infrastructure focused organizations, Fable 5 remains in a different category. That model was originally built for broader consumer use and was designed with stronger guardrails to prevent it from engaging with sensitive topics that could facilitate AI-enabled harm.
The contrast between the two models illustrates the broader tension the AI industry continues to wrestle with how to deliver powerful tools to those who need them most without creating new vectors for misuse.
The government’s growing role in AI oversight
Federal involvement in AI regulation has intensified significantly in recent months. An executive order issued by President Trump in early June directed agencies to strengthen cybersecurity defenses and establish a formal testing mechanism for advanced AI models before they reach the public. The specifics of that framework are still being worked out.
Anthropic has taken a proactive stance, dispatching a team of senior scientists and engineers to Washington, D.C., to work directly with the Commerce Department and the Office of the National Cyber Director. The goal is to build a pathway toward broader public access while keeping potential cyber risks in check.
OpenAI pushes back on staggered release approach
The timing of Anthropic‘s restoration overlaps with a notable moment for competitor OpenAI, which recently unveiled its GPT-5.6 family of models. OpenAI CEO Sam Altman has been vocal about his frustration with the government’s staggered release approach, calling it an obstacle to the company’s broader rollout plans. OpenAI has since committed to working with federal officials on a more structured vetting process for future model releases.
Both companies now find themselves navigating a regulatory environment that is moving faster and with more authority than many in the AI industry anticipated just a year ago.
