The National Security Agency is deploying Anthropic's newest and most capable AI model, Mythos Preview, even as the Defense Department that technically oversees it wages a legal campaign to brand the company a national security threat. Two sources confirmed the arrangement to Axios.
The disconnect exposes a fundamental tension within the U.S. government: operational cybersecurity needs colliding head-on with the Pentagon's institutional grievance against Anthropic. While the Defense Department pursues an active legal case to sever ties with the company, military and intelligence agencies are moving in the opposite direction.
The Pentagon launched its blacklist effort in February, demanding that Anthropic make its technology available for "all lawful purposes" and threatening vendors who work with the company. Anthropic refused, insisting on contractual walls against mass domestic surveillance and autonomous weapons development. That standoff remains unresolved in court.
The NSA's use of Mythos marks a significant escalation. Anthropic has restricted the powerful model to roughly 40 organizations, citing the severity of its offensive cyber capabilities. Only 12 of those recipients have been publicly named. The NSA gained access as an unnamed partner, according to one source, while another indicated broader adoption across the intelligence agency itself.
The specific applications remain unclear, but other organizations with Mythos access have primarily employed it to hunt for vulnerabilities within their own networks and systems. The model's offensive potential is why Anthropic has kept the release so tightly controlled.
Britain's National Cyber Security Centre has confirmed access to Mythos through the country's AI Security Institute, suggesting the arrangement extends beyond U.S. borders.
The timing of the NSA's deployment coincides with a high-level diplomatic push. Anthropic CEO Dario Amodei met Friday with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss how government agencies can integrate Mythos more broadly. Both parties characterized the conversation as productive, with attention expected to shift toward bringing departments outside the Pentagon into the fold.
The feud between the Pentagon and Anthropic emerged during contract renegotiations earlier in 2024. Some Defense officials continue to argue that the company's refusal to abandon safeguards proves Anthropic cannot be relied upon when military needs arise. Anthropic disputes this characterization. Other officials within the administration have simply decided they would rather have access to cutting-edge AI tools than continue the costly dispute.
Author James Rodriguez: "The NSA's move makes the Pentagon's legal case look increasingly hollow, and Washington's infighting over AI is finally breaking in favor of actually using it."
Comments