The National Security Agency (NSA) is reportedly employing Anthropic’s advanced AI model, Mythos Preview, despite a ban by the Department of Defense. This development underscores a significant contradiction within the U.S. national security framework, as highlighted in a recent report by Axios.
The Pentagon’s Ban on Anthropic
Back in July 2025, Anthropic secured a $200 million contract with the Department of Defense. The agreement included strict provisions against the use of its Claude AI for widespread domestic surveillance and autonomous weaponry. However, tensions arose in January 2026 when Defense Secretary Hegseth mandated the removal of these safety measures from all AI contracts, a move Anthropic opposed.
In response, the Pentagon labeled Anthropic as a ‘Supply-Chain Risk to National Security’ in February 2026, urging contractors to terminate business dealings with the company. Furthermore, President Trump instructed federal agencies to cease using Anthropic’s technology, with a phase-out period of six months for existing systems.
NSA’s Continued Use of Mythos
Despite the Pentagon’s prohibition, the NSA has continued to deploy Mythos Preview, Anthropic’s latest AI model. According to sources cited by Axios, the model is being utilized extensively within the agency. Currently, Mythos access is highly restricted, with only about 40 organizations having authorization, and the NSA is among those with undisclosed permissions.
The advanced offensive cybersecurity capabilities of Mythos have prompted Anthropic to limit its availability. This secrecy highlights concerns over the potential for the AI to enhance cyberattacks if misused, making the NSA’s adoption noteworthy given its role in cybersecurity and signals intelligence.
Ongoing Legal Disputes and Future Implications
In March 2026, Anthropic initiated legal proceedings in San Francisco, arguing that the Pentagon’s classification was ‘unprecedented and unlawful,’ infringing on free speech and due process rights. While the case progresses, the military contends that Anthropic’s tools pose national security risks while using them internally.
On April 17, Anthropic CEO Dario Amodei met with White House officials to discuss government use of Mythos and potential expansion plans. The NSA’s clandestine use of Mythos, amidst ongoing legal battles, suggests operational needs are taking precedence over official policies within U.S. intelligence.
As AI technology evolves, the disparity between institutional bans and practical use is widening. This raises important questions about governance, oversight, and accountability in deploying advanced AI within national security frameworks.
