White House Drafting Guidance to Restore Anthropic Access

The White House is drafting guidance to restore federal agency access to Anthropic’s AI systems, including its Mythos model, even though the Department of Defense labeled the company a “supply chain risk” in February and directed agencies to stop using its products.

The guidance under consideration would create a controlled path for agencies to use Anthropic technology for cybersecurity and defense purposes. Administration officials have met with Anthropic and industry leaders to discuss limited federal use of Mythos.

Anthropic CEO Dario Amodei met earlier this month with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. Officials also convened meetings with several major banks — including Citigroup, Bank of America, Wells Fargo, Morgan Stanley and Goldman Sachs — to review cybersecurity risks tied to Mythos.

Federal departments and private security teams have pursued access to Mythos because the model can identify and, in testing, exploit software vulnerabilities across major operating systems. The National Security Agency is testing a preview of Mythos on classified networks.

In one evaluation, the model completed a 32-step corporate network attack simulation in less time than a typical human team. A browser developer reported that Mythos-assisted testing identified 271 vulnerabilities that were then patched.

Anthropic limited public release of Mythos, offering access mainly to governments and large technology companies so they could assess the model’s capabilities and use it to strengthen software security.

The Pentagon designated Anthropic a supply chain risk after the company declined to give the department unrestricted access to its models and set policies restricting use for surveillance and autonomous weapons development.

Former President Donald Trump posted on Truth Social in February that the United States would “never allow a radical left, woke company to dictate how our great military fights and wins wars!” He wrote that decisions about military action belong to the commander-in-chief.

Company and government representatives are negotiating technical and policy controls that could permit use of Anthropic models without removing all of the company’s operational limits. Mozilla, which tested Mythos, wrote, “For a hardened target, just one such bug would have been red-alert in 2025, and so many at once makes you stop to wonder whether it’s even possible to keep up.”

Anthropic did not immediately respond to requests for comment. The draft guidance seeks to balance access for federal agencies with the security concerns the Pentagon raised.