NSA uses Anthropic's classified Mythos model despite conflict with the Pentagon

The NSA secretly works with Anthropic's classified AI model Mythos — and this occurs despite the company's public confrontation with the Pentagon. According to TechCrunch, the U.S.

National Security Agency has already integrated the limited model into its intelligence operations, despite officially strained relations between the developer and the military department. The publication's sources claim that it is not a pilot test, but actual deployment. Mythos is a limited Anthropic model that is not available to the general public.

Unlike Claude, which is distributed through API, subscription, and corporate contracts, Mythos exists in a separate, closed circuit. Details about the model's capabilities, training methods, system instructions, and licensing terms are not publicly disclosed. The very existence of Mythos had not been officially confirmed by Anthropic until recently.

Analysts suggest several versions of how Mythos differs from standard Claude. Possible explanations: expanded context windows for working with long analytical documents, additional accuracy when processing specific intelligence report formats, relaxed restrictions on certain types of queries relevant to the government sector. An alternative explanation: "limited" simply means controlled distribution — the model is not sold openly, but supplied only to verified government partners.

Anthropic consistently positions itself as a "responsible" AI developer: the company promotes principles of safe technology development and declares a cautious approach to military applications. This has created a noticeable public distance between Anthropic and the defense establishment — unlike, for example, Palantir, Anduril, or Scale AI, which openly build their business on contracts with the Department of Defense. Nevertheless, the line between "civilian" and "government" AI providers turns out to be thinner than commonly believed.

The intelligence community operates in a different framework than the Pentagon: the NSA reports to the Director of National Intelligence, not the Department of Defense. Formally, this allows the agency to enter into contracts with suppliers independently of the military's position — even if publicly a company distances itself from "military" applications. The integration of AI into the work of U.

S. intelligence services is nothing new. The CIA launched an internal tool based on large language models back in 2023, and by 2025 expanded cooperation with several AI companies.

The NSA traditionally stays ahead of public discourse in the application of new technologies. For intelligence structures, key AI applications are processing vast arrays of intercepted signals, automating pattern search in communications, and accelerating analytical reports. In this context, the emergence of a powerful language model in the agency's arsenal is more a logical step than sensational.

The situation places Anthropic before a contradiction familiar to the entire technology sector: public ethical positions and actual commercial practices do not always coincide. For a company actively building an image of a "safe" developer in contrast to more aggressive market players, reports of collaboration with the country's premier intelligence agency represent a reputational challenge. Neither Anthropic nor the NSA officially commented on the publication's information.

While the public narrative about the company's conflict with the Pentagon lives its own life, the actual business relations with the intelligence community apparently proceed as usual.