Anthropic's CEO met with the White House to discuss government access to its Mythos AI, a tool for finding software bugs, signaling high-level interest in its technology.
The meeting occurs amid tension, as Anthropic is simultaneously challenging a rare 'supply chain risk' designation from the Pentagon, creating a complex government dynamic.
Anthropic is selectively sharing Mythos with tech giants like Apple and Amazon, but keeps it private due to fears it could be weaponized to exploit vulnerabilities.
Atlas AI
Anthropic CEO Dario Amodei has held discussions with White House Chief of Staff Susie Wiles about potential government access to the company’s advanced cybersecurity-focused AI model, known as Mythos. Officials and the company have described the talks as focused on how U.S. government agencies could use the tool to identify and surface software vulnerabilities.
The outreach comes as Anthropic faces a separate, high-stakes dispute with the U.S. Defense Department. The company is challenging a Pentagon designation that labels Anthropic a supply chain risk, a classification described as rare and severe for a U.S.-based firm. The parallel tracks—engagement with the White House while contesting a Defense Department determination—highlight the complicated posture of the company inside the U.S. national security ecosystem.
Anthropic has positioned Mythos as a model built to find weaknesses in software that can remain undetected for long periods. The company has pointed to a case in which Mythos identified a 27-year-old bug in an operating system commonly used in internet routers and firewalls. Anthropic has said it does not plan to release Mythos publicly, citing concerns that broad availability could enable malicious actors to weaponize the same vulnerability discovery capabilities to attack systems.
Instead of a public release, Anthropic has shared Mythos with a limited private-sector group of over 40 technology companies. That coalition includes Apple, Amazon, and Microsoft, and it has adopted the name “Project Glasswing”. The label references the glasswing butterfly, whose transparent wings help it blend into its surroundings, a metaphor the group uses for Mythos’s ability to uncover long-hidden flaws inside complex codebases.
In parallel with the private-sector effort, the White House meeting signals that parts of the U.S. executive branch are exploring whether Mythos could be applied to public-sector cybersecurity needs. Mythos has reportedly found thousands of bugs across widely used software programs used by millions, and proponents of agency access argue that earlier detection could strengthen defenses by identifying weaknesses in critical systems before they are exploited.
What remains unclear is whether the discussions will lead to a formal agreement, and how any arrangement would be structured given the ongoing Pentagon dispute. The outcome could influence how government agencies work with leading AI firms on sensitive security tooling, while also intersecting with Anthropic’s effort to contest the Defense Department’s supply chain risk designation.
