NSA Deploys Anthropic's Mythos AI Despite Pentagon Blacklist

NSA Embraces Banned AI as Cybersecurity Needs Trump Pentagon Feud

In a striking contradiction of U.S. defense policy, the National Security Agency (NSA) is actively using Anthropic's most powerful and restricted AI model, Mythos Preview, according to multiple sources familiar with the matter. This deployment occurs even as the Department of Defense (DoD), which oversees the NSA, maintains that Anthropic presents a "supply chain risk" to national security and is engaged in ongoing litigation to block its use.

The situation underscores a deep and public rift within the national security establishment. On one side, operational agencies like the NSA prioritize cutting-edge tools for cyber defense. On the other, Pentagon procurement officials are locked in a legal battle with an AI company over contractual terms and perceived reliability.

The Blacklist and the Breach

The conflict originated in February 2026, when the Department of Defense moved to cut off Anthropic and force its vendors to follow suit. The dispute stemmed from tense contract renegotiations where the Pentagon demanded Anthropic's Claude model be available for "all lawful purposes." Anthropic refused, insisting on walling off applications for mass domestic surveillance and autonomous weapons development.

Some Defense officials viewed this refusal as proof the company "can't be trusted to be there when the military needs it," a stance Anthropic denies. The DoD subsequently declared Anthropic a supply chain risk—a historic designation previously reserved for foreign companies—and initiated the blacklisting process. That case remains ongoing.

Despite this, two sources confirmed to Axios that the NSA is using Mythos. A third source indicated the model is seeing broader, albeit discreet, use within the Pentagon itself. This suggests the imperative for advanced cybersecurity tools is overriding the department's own policy directives.

Mythos: A Restricted Cyber Weapon

Anthropic's Mythos Preview is not a publicly available model. Due to concerns over its powerful offensive cyber capabilities, the company restricted initial access to approximately 40 select organizations. Anthropic has publicly named only 12, including corporate giants like JPMorgan, Amazon, and Apple, stating the goal is to help them find critical code flaws before state-sponsored hackers do.

The NSA is among the unannounced organizations granted early access. It's unclear precisely how the agency is deploying Mythos, but other authorized entities primarily use it to scan their own IT environments for exploitable security vulnerabilities—a defensive application. The United Kingdom's intelligence services have also publicly stated they have access to Mythos through the country's AI Security Institute.

The model's perceived power is central to the dilemma. One administration official, speaking to Axios, accused Anthropic of using "fear tactics" by highlighting how Mythos could supercharge hacking, calling it a "cyber weapon" used to find "friendly ears in the government."

White House Seeks a Thaw, Pentagon Stays Cold

Parallel to the NSA's operational use, high-level negotiations are underway to normalize Anthropic's relationship with the broader federal government. On April 17, 2026, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss Mythos's use within government and Anthropic's security practices.

Sources described the meeting as "productive," with next steps focusing on how departments outside the Pentagon might engage with the model. The Office of Management and Budget has reportedly sent emails to agencies inquiring about Mythos, indicating a coordinated review. "There's progress with the White House. There's not progress with [the Department of] War," one administration official starkly summarized.

This schism is highlighted by reports that staff from at least two other large federal agencies have reached out to Anthropic for Mythos access despite the ban, and multiple congressional committees have requested briefings.

Why This Strategic Split Matters

The NSA's use of Mythos despite the blacklist is more than bureaucratic inconsistency; it's a signal of the overwhelming pressure facing U.S. cyber defenders. The agency's primary mission includes securing government networks and gathering signals intelligence, both arenas where AI capable of finding software vulnerabilities at scale could be a decisive advantage.

This creates a classic capability versus control conflict. The Pentagon's procurement and legal arms prioritize contractual control, supply chain security, and unambiguous availability for all military scenarios. Frontline cyber operators, however, prioritize having the most effective tool immediately, even if its provider imposes ethical use restrictions.

The situation also exposes the challenges of governing "dual-use" AI technology. Mythos, designed to defend networks by finding holes, can equally be used to attack them. Restricting it from a government with legitimate defense needs is impractical, yet integrating it comes with strings attached from a private company.

Ultimately, the NSA's actions suggest that in the high-stakes realm of cybersecurity, operational necessity is forcing a pragmatic—if legally messy—path forward. The White House talks indicate a political desire to resolve the impasse, but the Pentagon's court case remains a significant barrier. The U.S. government is, for now, simultaneously suing Anthropic and using its most advanced product, a paradox that speaks to the disruptive and urgent nature of frontier AI.