The National Security Agency (NSA) has informed lawmakers that it is spending billions of dollars in taxpayer funds this year on the evaluation and testing of advanced artificial intelligence models, according to two sources familiar with classified intelligence estimates. This expenditure is significantly higher than previously known, leading lawmakers to believe that a comprehensive AI regulatory system could cost the government tens of billions of dollars annually.
President Donald Trump has largely resisted calls for increased federal oversight of AI, rejecting regulatory proposals from Congress and frontier labs aimed at imposing new safeguards on AI models. However, following a series of high-profile hacking and security incidents, the NSA’s Artificial Intelligence Security Center has begun testing frontier models to identify potential national security vulnerabilities.
The high costs associated with these efforts are intensifying discussions about the government's role in reviewing AI models developed by major technology companies. There are growing calls for frontier labs to bear the financial responsibility for reviewing these models, especially as spending is expected to increase with advancements in technology.
Previous proposals for establishing federal AI regulatory oversight suggested a much lower cost. For example, the Congressional Budget Office estimated that the AI Security and Innovation Act, a bipartisan proposal to create a center focused on AI risks, would cost approximately $20 million per year. Another House bill proposing a new reporting and tracking system for AI has a total cost of $36 million over five years, according to the CBO.
Some technology leaders, including billionaire Elon Musk, have proposed that frontier labs should evaluate each other’s models before release without government oversight. Recently, it was reported that Google, OpenAI, and Anthropic are collaborating on a plan to establish their own AI safety-focused standards body. However, some AI safety experts have criticized this approach as allowing firms to self-regulate. An alternative proposed is for the government to impose a tax on AI companies to fund these safety operations, ensuring independence and preventing the financial burden of regulation from falling primarily on taxpayers. Both Anthropic and OpenAI have expressed a desire for greater federal oversight and may be willing to contribute financially.
Currently, the NSA’s AI testing initiatives are funded through classified segments of the federal national security budget, although the exact amount spent to date remains unclear. The Pentagon has declined to comment on the NSA's AI expenditures, citing security reasons. A spokesperson for the Defense Department stated, "For security reasons, the Department does not discuss the technical architecture or resource allocation for its AI tools."
The Defense Department’s annual budget is nearly $1 trillion, and some funds already allocated to the military could potentially be redirected towards AI-related spending, which may lower the overall costs.
One source indicated that a significant portion of the NSA's AI-related expenses has been attributed to the high costs of computing power needed for running and testing AI models. The global demand for computing power has surged, leading to increased prices amid insufficient chip production. For instance, Anthropic has secured computing deals that could reach costs of up to $517 billion, according to The Information.
Another major expense involves hiring personnel, as top AI engineers are being offered salaries in the hundreds of millions of dollars, far exceeding government pay scales. Despite the NSA being recognized for its technical expertise, it may struggle to compete with the lucrative compensation packages offered by frontier labs.
Nathan Calvin, general counsel at Encode, an AI advocacy organization, stated, "Having in-house AI evaluation capability I think is extremely important and necessary. But it is genuinely expensive. You’re competing in bidding with some of the most price-insensitive customers."
Experts have urged lawmakers to expedite efforts to enhance third-party audits of AI models. Nat Purser, director of U.S. Policy at the AI Verification and Evaluation Research Institute, noted that while "people often underestimate how much computing power it can take to rigorously test advanced AI systems," the government should invest in the necessary resources. Purser also suggested that AI companies could bear the costs if taxpayers are contributing significant amounts to evaluate the models. "If we want the government to be equipped to assess these systems, we need to fund the computing resources and expertise that requires," Purser said. "I think of these tests as public goods, but there are reasonable questions about if taxpayers should foot the bill. An assessment on frontier developers could be required to help fund these independent audits, including the necessary computing resources."