Classified Estimates Put NSA AI Model Testing Costs in the Billions

Summary

The National Security Agency told lawmakers it is spending billions of dollars this year to evaluate advanced AI models, according to two people familiar with classified intelligence estimates. The reported figure is much higher than earlier federal cost projections and has increased pressure for a clearer system of government oversight. The NSA’s Artificial Intelligence Security Center began testing frontier models after high-profile hacking and security incidents, with the aim of identifying national security vulnerabilities. Earlier Congressional Budget Office estimates put proposed federal AI oversight programs at roughly $20 million a year or $36 million over five years, but the NSA’s current spending suggests that operational testing can be far more expensive. The exact amount spent remains unknown because the work is funded through classified portions of the national security budget, and the Pentagon declined to discuss its allocation. Sources said computing power is the largest cost, as testing requires substantial chip capacity during a global shortage; hiring specialized personnel is another major expense because government salaries cannot match some frontier-lab offers. The article describes competing proposals for paying for independent evaluations: continued taxpayer funding, a tax or assessment on frontier AI developers, or industry-led reviews. Supporters of government capability argue that independent testing is necessary, while critics warn that companies should not effectively police themselves. Anthropic and OpenAI have indicated support for greater federal oversight and may be willing to help pay for it, while Google, OpenAI, and Anthropic are reportedly working on a private AI safety standards body. The article says spending could grow as models become more advanced, but it does not establish the exact current total or a final regulatory plan.