NSA Continues Using Anthropic Models Despite Official Warnings

The U.S. National Security Agency continues to use artificial intelligence from Anthropic — despite American authorities declaring the company a potential threat to national security.

While the federal government seeks answers about how to manage its relationship with this developer, security agencies operate by their own logic. The situation looks frankly paradoxical. On one hand, Anthropic — creator of Claude, one of the most powerful commercial AI assistants — has come under the scrutiny of American regulators.

On the other hand, the NSA, whose work is tied to processing vast quantities of intelligence data, has clearly found applications for these tools and has no intention of abandoning them. There have been no official comments about what specific tasks the models are applied to — but the logic is obvious: document summarization, linguistic analysis, automation of routine analytical operations. At the center of complaints against Anthropic are several factors.

The company is registered in the United States, however among its major investors is Google, and its founders come from OpenAI. This alone does not make it a threat, but against the backdrop of mounting technological confrontation with China, American regulators have become more scrutinizing of any foreign connections in the AI sector. Anthropic's negotiations about expanding partnerships with Asian investors apparently attracted heightened attention from authorities.

This is not the first time U.S. government agencies have faced such internal contradictions.

The CIA tested its own AI platform based on large language models back in 2023, the Pentagon actively implemented generative AI into analytical processes. The difference is that then it concerned closed, isolated environments — not publicly available commercial products, whose status is now being disputed at the regulatory level. It is important to understand: the use of commercial AI in government agencies is not in itself extraordinary.

The U.S. has long integrated civilian technologies into federal agencies' operations — special certification programs like FedRAMP exist for this purpose.

Another question is what happens when an agency continues to work with a tool that its colleagues from another branch of government declare risky. It is precisely this gap between official rhetoric and actual practice that makes the NSA situation instructive. The regulatory body sees the risk and documents it.

The operational agency sees a tool that works, and continues to use it. Until a unified policy is developed, both positions exist in parallel — and no one is in a hurry to reconcile them. For Anthropic itself, the situation is ambiguous.

Government contracts are stable revenue and a sign of institutional trust. But being at the center of a discussion about national security is not something most technology companies would want for their brand. Especially in conditions of fierce competition with OpenAI and Google DeepMind, which are also competing for government contracts.

How the story will end with the assessment of Anthropic as a source of risk remains unclear. But the precedent is noteworthy: it shows that even with declared caution, the state remains dependent on specific technology players. And breaking this dependence is not simple — even when political will appears to do so.