Spy Agencies Use Anthropic’s Mythos Despite DoD Blacklist

U.S. intelligence agencies have deployed Anthropic’s Mythos Preview in operational settings to scan networks and automate complex system analysis for cybersecurity even after the Defense Department designated the company a supply-chain risk in February 2026.

Officials involved in the deployments say Mythos is used to locate deeply embedded software bugs and to simulate or suggest methods to exploit those bugs for testing and remediation. The model’s autonomous analysis capabilities are a primary reason some units prioritize its use for defense work.

Access to Mythos is limited to about 40 vetted organizations; roughly a dozen of those users are publicly identified, while several remain unlisted, including a major intelligence agency reported to be the National Security Agency. Agencies in the United Kingdom also have access through their national AI security institute.

The Pentagon’s blacklist bars Anthropic tools from standard Defense Department use and followed a dispute over required safeguards and military applications. Some national security officials argue that restricting access to a powerful capability could leave the U.S. at a disadvantage compared with foreign rivals. Other defense officials point to Anthropic’s refusal to embed its AI in certain military systems and describe the firm as a potential “flight risk” if it could cut off services during a crisis.

Anthropic’s CEO, Dario Amodei, has held talks with government officials and confirmed the company is open to collaboration. White House officials met with Amodei; a White House statement read, “We discussed opportunities for collaboration, as well as shared approaches and protocols to address the challenges associated with scaling this technology.”

Anthropic filed suit to overturn the supply-chain designation, calling the label punitive after it declined a Defense Department request to integrate the model into fully autonomous weapons systems and mass domestic surveillance. A California district court temporarily blocked the designation, but a federal appeals court later lifted that stay and the blacklist remains in effect.

During the early blacklist effort, President Donald Trump criticized Anthropic, saying, “We don’t need it, we don’t want it, and will not do business with them again!”

Officials and cybersecurity experts note Mythos’ ability to find and autonomously analyze vulnerabilities could be misused if it fell into hostile hands. Intelligence units using the model say strict vetting and access controls are in place to limit misuse.

The continued operational use of Mythos by intelligence agencies, despite the Pentagon’s formal restrictions, reflects an ongoing policy challenge for federal agencies in adopting advanced AI tools for defense while managing supply-chain and control risks.