NSA tells lawmakers it spends billions a year testing AI models

The National Security Agency has told lawmakers that it is spending billions of dollars of taxpayer money this year on evaluating and testing advanced artificial intelligence models, according to two sources familiar with classified intelligence estimates.

The price tag is significantly higher than previously known, and lawmakers now believe that a comprehensive system of AI regulation could cost the government tens of billions of dollars a year. The testing is paid for out of classified parts of the national security budget, and the exact amount spent so far is unclear.

The Pentagon declined to comment on the spending. "For security reasons, the Department does not discuss the technical architecture or resource allocation for its AI tools," a Defense Department spokesperson said.

Why the testing began

The NSA's Artificial Intelligence Security Center started checking frontier models after a string of high-profile hacking and security incidents, looking for vulnerabilities that matter for national security. President Donald Trump has mostly resisted calls for greater federal oversight of AI.

Compute is the biggest cost

According to one source, the largest AI-related cost so far has been computing power: the chip processing needed to run and test the models. Anthropic has lined up computing deals worth as much as $517 billion, according to The Information.

Hiring is the other big expense: some top AI engineers have been offered salaries in the hundreds of millions of dollars, which the government cannot match. "Having in-house AI evaluation capability is extremely important and necessary, but it is genuinely expensive," said Nathan Calvin, general counsel at the AI advocacy group Encode.

Who should foot the bill

Earlier proposals for federal AI oversight carried far smaller estimates: the Congressional Budget Office put the bipartisan AI Security and Innovation Act, which would create a centre on AI risks, at about $20 million a year.

Billionaire Elon Musk has suggested that the frontier labs should review one another's models before release instead of answering to government. According to The Information, Google, OpenAI and Anthropic are working on a plan for their own AI safety "standards body"; some safety experts call that self-policing and propose a levy on AI companies instead, so the burden does not fall mainly on taxpayers.

Nat Purser, director of US policy at the AI Verification and Evaluation Research Institute, said people often underestimate how much computing power rigorous testing of advanced systems takes. "An assessment on frontier developers could be required to help fund these independent audits, including the necessary computing resources," he said.