NSA deploys Anthropic's Mythos AI model for cyber operations despite Pentagon blacklist

NSA Deploys Anthropic Engineers for Mythos AI Implementation

The NSA has embedded approximately half a dozen Anthropic engineers inside the agency to help deploy and customize the Mythos AI model for specialized applications, according to a Financial Times report citing anonymous sources 1

1

. These Anthropic engineers embedded inside the agency are tasked with guiding the intelligence organization's use of the cybersecurity AI model, though it remains unclear whether they are actively participating in live hacking operations 2

2

.

The deployment supports offensive cyber operations aimed at infiltrating networks in countries such as China and Iran, according to people familiar with the arrangement 3

3

. The rationale behind this move centers on the expectation that U.S. adversaries will likely deploy similar cybersecurity capabilities, making early advantage critical in an evolving threat landscape.

Pentagon Blacklist Creates Contradictory Government Stance

This collaboration occurs despite Anthropic being designated a supply chain risk by the Department of Defense in early March 2026, placing the American company in the same category as Huawei and ZTE 2

2

. The dispute originated during negotiations over a $200 million contract when the Trump administration demanded that Anthropic allow usage of its technology for "all lawful purposes," which would have removed AI guardrails protecting against mass domestic surveillance and autonomous weapons 2

2

.

Anthropic CEO Dario Amodei refused to comply, posting a statement defending the company's position that "in a narrow set of cases, [Anthropic believes] AI can undermine, rather than defend, democratic values" 2

2

. The company subsequently filed a lawsuit against the Department of Defense, arguing the designation violated the First Amendment . Two separate courts have issued contradictory rulings on the matter, with a California federal judge temporarily blocking one version of the designation in March, while a Washington appeals court allowed the Pentagon blacklist to remain in place in April 3

3

.

Federal Ban Doesn't Stop Intelligence Gathering Applications

An earlier Axios report from April revealed that the NSA was already using Mythos despite a federal ban on using Anthropic's technology 1

1

. The NSA's mission includes collecting intelligence from wiretaps, undersea cables, corporate partnerships and other clandestine means, as well as conducting offensive cyberattacks on foreign adversaries 1

1

. This suggests that parts of the Department of Defense may consider the Mythos AI model too valuable to ignore for AI for national security purposes, even as the Pentagon argues in court that Anthropic's technology poses a national security risk 3

3

.

By May, the Pentagon had signed major AI deals with OpenAI, Google, Microsoft, and Nvidia while excluding Anthropic 3

3

. Ironically, after public backlash, OpenAI added safety clauses to its deal, which the Department of Defense accepted, suggesting that the dispute with Anthropic may have been politically motivated 2

2

.

Mythos Expands Access Through Project Glasswing

Anthropic has positioned Mythos as a defensive cybersecurity tool, launching Project Glasswing in April with partners including AWS, Google, Microsoft, Nvidia, and CrowdStrike 3

3

. The company announced that Mythos had identified thousands of zero-day vulnerabilities, with partners finding more than 10,000 high- or critical-severity flaws 3

3

. Anthropic claims it had to limit access to the model initially, fearing its cybersecurity capabilities could be exploited to discover security flaws and carry out hacks 1

1

.

The Mythos AI model is currently in early-access stage, with Anthropic expanding distribution to approximately 150 organizations across 15 countries, up from only a handful of U.S. organizations after its April release 2

2

. Reports suggest the NSA was part of the smaller initial group that received access to the model 4

4

. When reached for comment, an NSA spokesperson declined to confirm or deny the reporting, while Anthropic did not respond to requests for comment 1

1

.