The U.S. National Security Agency has secured access to Anthropic’s restricted Mythos AI model, despite Pentagon opposition. The move highlights growing internal conflict over deploying advanced AI tools with potential offensive cyber capabilities.
Two sources told Axios that the NSA is among a select group granted access to Mythos Preview, a model Anthropic has tightly controlled due to its risk profile. The Pentagon had moved in February to cut ties with the company and instructed vendors to follow, citing national security concerns tied to misuse scenarios.
Why Is The Pentagon Blocking Anthropic While Agencies Adopt It?
The dispute escalated after the Pentagon pushed for unrestricted access to Anthropic’s Claude model for “all lawful purposes.” Anthropic resisted, seeking limits on applications such as mass surveillance and autonomous weapons. This disagreement has led to a legal standoff, with defense officials questioning the firm’s reliability even as other agencies expand usage.
Government demand for Mythos centers on cybersecurity applications. Of roughly 40 organizations with access, most use the model to identify vulnerabilities within their own systems, according to sources. Anthropic has publicly disclosed only a portion of these partners, including the U.K.’s AI Security Institute, leaving several U.S. agencies undisclosed.
Anthropic has denied claims that its tools pose unmanaged risks in military contexts, while maintaining restrictions on high-risk use cases. The company’s leadership recently met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss broader government deployment, with sources describing the talks as productive.
Still, the divergence between Pentagon policy and agency-level adoption suggests fragmented governance over advanced AI systems. The next catalyst will be whether the administration formalizes a unified framework for deploying high-risk AI across defense and civilian agencies.
