Two people familiar with classified estimates say the NSA told lawmakers it is spending billions this year to test frontier AI models. No public budget document confirms the amount, and the Pentagon has not commented.
The reported spending is tied to the NSA's Artificial Intelligence Security Center, created to examine frontier models for national-security vulnerabilities, but no official NSA, Defense Department or congressional source has confirmed that link or the evaluations' scope. The reporting identifies computing infrastructure and specialist personnel as key costs; no itemized NSA budget is public.
By comparison, the Congressional Budget Office estimated H.R. 9363 would authorize $20 million annually for a civilian NIST center measuring AI risks in fiscal years 2027–32, with $80 million in implementation costs estimated for 2026–31; the aims differ. Lawmakers have reportedly considered developer fees for independent evaluations, but available sources do not establish that Anthropic or OpenAI agreed to pay. Anthropic, Google and OpenAI have also discussed a safety-standards body, which OpenAI confirmed. OpenAI says industry standards should complement federal safeguards; no public charter, governance or funding plan had been identified at the time of reporting.
