White House Plans Controlled Access to Anthropic’s Mythos AI Model for Federal Agencies

The White House is preparing to provide select federal agencies with access to a version of Anthropic’s advanced AI model, Mythos, as part of a tightly controlled initiative aimed at evaluating its cybersecurity capabilities.

The move follows the April 7 announcement of “Project Glasswing,” under which a limited group of organisations are granted access to the unreleased Claude Mythos Preview model for defensive cybersecurity applications. The initiative reflects growing interest in leveraging frontier AI systems to identify and mitigate vulnerabilities across critical infrastructure.

According to reports, Mythos has already demonstrated the ability to uncover thousands of significant vulnerabilities across operating systems, web browsers, and other software environments. Its advanced coding capabilities enable it not only to detect weaknesses but also to model potential exploitation pathways—raising both opportunities and risks for cybersecurity operations.

Gregory Barbaccia, based at the Office of Management and Budget, indicated that the government is working to establish safeguards before broader deployment. In communications to federal agencies, Barbaccia emphasized coordination with AI providers, industry partners, and intelligence agencies to ensure appropriate controls are in place.

While no definitive timeline has been confirmed, the initiative signals a cautious but strategic approach to integrating cutting-edge AI into government cybersecurity frameworks. The potential deployment comes amid heightened concerns that such powerful models could be misused if not properly governed.

Anthropic has been actively engaging with U.S. policymakers on the implications of Mythos, even as its relationship with the United States Department of Defense has faced challenges following a reported contract dispute.

As governments worldwide assess the risks and benefits of frontier AI, the White House’s move underscores a broader trend toward controlled experimentation—balancing innovation in cybersecurity with the need for strict oversight and risk mitigation.