Classified Estimates Show the NSA Is Paying Billions to Test AI Models
Classified estimates say the NSA is spending billions this year testing frontier AI models for security risks.
Sources familiar with classified estimates say the NSA is spending billions of dollars this year evaluating advanced AI models, leading lawmakers to fear broader federal oversight could cost tens of billions annually. Earlier CBO scores put related proposals far lower: about $20 million a year for one AI risk center bill and $36 million over five years for a reporting system. The NSA Artificial Intelligence Security Center began testing frontier models after high-profile hacking and security incidents, with compute and hiring as the main costs. Google, OpenAI, and Anthropic are reportedly discussing their own safety standards body, which some experts reject as self-policing.
- NSA frontier-model testing is reportedly costing billions this year.
- Earlier CBO scores for related AI bills were only tens of millions.
- Compute purchases and specialist hiring are the largest expenses.
- Google, OpenAI, and Anthropic may form a safety standards body.
- Critics say industry self-review would let labs police themselves.
Full article821 words · extracted from washingtonsun.com · click to collapse
The National Security Agency told lawmakers that it is spending billions of dollars in taxpayer funds this year evaluating and testing advanced artificial intelligence models, according to two sources familiar with classified intelligence estimates.
The price tag — which is significantly greater than previously known — has led lawmakers to believe that a more comprehensive AI regulatory system could cost the government tens of billions of dollars per year, the sources said.
President Donald Trump has mostly resisted calls for greater federal oversight of AI, rejecting regulatory proposals from members of Congress and the frontier labs aimed at imposing new safeguards on the models. But after a string of high-profile hacking and security incidents, the NSA’s Artificial Intelligence Security Center began testing frontier models to identify potential national security vulnerabilities.
The high cost of those efforts is bringing new urgency to the debate about the government’s role in reviewing AI models designed by some of the most powerful and resourceful companies in the world. It also may intensify calls to have the frontier labs to shoulder the financial burden of reviewing the models, especially with spending set to only balloon as the technology grows more advanced.
Estimates of prior proposals to establish new federal AI regulatory oversight suggested a far lower price tag.
For instance, the Congressional Budget Office estimated the AI Security and Innovation Act, a bipartisan House proposal to establish a center on AI risks and “facilitate the mitigation of those risks,” would cost roughly $20 million per year.
A separate bill in the House to establish a new reporting and tracking system for AI calls for $36 million in total over the next five years, the CBO reported.
Some tech leaders, like billionaire Elon Musk, have suggested that the frontier labs should review each other’s models prior to their release without government oversight. On Thursday, The Information reported that Google, OpenAI and Anthropic are jointly working on a plan to create their own AI safety-focused “standards body.”
But some AI safety experts have rejected this model as effectively allowing the firms to police themselves. One alternative is for the government to levy a tax on AI companies to run these safety operations — both to ensure independence and to ensure the financial burden of regulating AI does not fall primarily on the taxpayers. Anthropic and OpenAI have suggested that they want greater federal oversight and may be open to paying for it as well.
Currently, NSA’s AI testing efforts are being funded by classified portions of the federal national security budget, the two sources familiar said. The exact dollar amount the government has spent so far is not clear.
The Pentagon declined to comment on what the NSA is spending on AI.
“For security reasons, the Department does not discuss the technical architecture or resource allocation for its AI tools,” a spokesperson for the Defense Department said.
The Defense Department’s annual budget is close to $1 trillion. Some budgetary funds already allocated to the military could be shifted into AI-related spending, reducing the potential price tag.
One source said the biggest AI-related cost to the NSA thus far has been the extra computing power: the processing on chips necessary to run and test the AI models. Purchasing computing power has become incredibly expensive worldwide amid a surge in demand accompanied by insufficient production of chips. For instance, Anthropic has lined up computing deals that could cost as much as $517 billion, according to The Information.
Another big expense is in hiring personnel. Some top AI engineers have been offered salaries in the hundreds of millions of dollars, far outpacing what the government could match. NSA is widely regarded as having the deepest roster of technical experts within the government and still may not be able to compete with the giant pay packages thrown around by the frontier labs.
“Having in-house AI evaluation capability I think is extremely important and necessary. But it is genuinely expensive,” said Nathan Calvin, general counsel at Encode, an AI advocacy organization. “You’re competing in bidding with some of the most price-insensitive customers.”
Some experts have pushed for lawmakers to swiftly move to strengthen third-party audits of the models. Nat Purser, director of U.S. Policy at the AI Verification and Evaluation Research Institute, said that while “people often underestimate how much computing power it can take to rigorously test advanced AI systems,” the government should make the investments necessary. Purser also said the AI companies could foot the bill if taxpayers are paying large sums 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.”