The National Security Agency is reportedly using Mythos Preview, the most advanced AI model developed by Anthropic, even as the United States Department of Defense has labeled the company a potential “supply chain risk.”
Conflicting Signals Within the Pentagon
According to reports, the NSA has secured access to the high-powered model, which Anthropic has kept tightly restricted due to its potential offensive cyber capabilities. This comes despite the Pentagon’s earlier move in February to distance itself from the company and instruct vendors to do the same.
The situation highlights a growing contradiction: while defense officials argue in court that Anthropic’s tools could pose national security risks, internal demand for the technology appears to be driving continued adoption across military-linked agencies.
Dispute Over AI Use Limits
Tensions escalated during contract renegotiations earlier this year. The Pentagon reportedly pushed for unrestricted access to Anthropic’s AI systems for “all lawful purposes.”
However, Anthropic resisted, seeking clear limitations—particularly around mass domestic surveillance and autonomous weapons development. This disagreement led some defense officials to question the company’s reliability in sensitive military contexts, a claim Anthropic has denied.
Why the Technology Is in Demand
The appeal of Mythos lies in its specialized capabilities. Organizations with access to the model primarily use it to identify vulnerabilities within their own digital infrastructure, strengthening cybersecurity defenses.
While Anthropic has publicly disclosed only a handful of partners—including the U.K.’s AI Security Institute—sources indicate that the NSA is among a broader, undisclosed group granted access.
Talks Move Beyond the Pentagon
Efforts to resolve the standoff have expanded beyond the Defense Department. Dario Amodei, CEO of Anthropic, recently met with senior U.S. officials, including Susie Wiles and Scott Bessent, to discuss broader government use of the technology.
Sources described the discussions as constructive, suggesting the administration may explore pathways to deploy Mythos across other federal agencies—potentially sidestepping the Pentagon’s current restrictions.
A Growing AI Policy Divide
The episode underscores a widening gap between AI innovation and regulatory caution within the U.S. government. As agencies weigh the strategic advantages of advanced AI against potential risks, cases like this are likely to shape how emerging technologies are governed in national security contexts.
