The NSA is reportedly spending billions of dollars just to test AI models, before it even puts them to work.
The Washington Sun reported on September 25, 2026, that classified budget estimates it reviewed show the National Security Agency has set aside billions of dollars specifically to evaluate artificial intelligence models. The figures come from internal estimates rather than a public contract announcement, and the report does not identify which companies' models are being tested or how the money breaks down between evaluation, licensing, and infrastructure. No dollar figure more precise than "billions" has been made public. The story picked up 29 points and 10 comments on Hacker News.
Government AI contracts are usually announced with press releases and dollar figures attached, but this spending sits entirely behind a classification wall, with no vendor named and no total confirmed. That suggests evaluating AI models for intelligence work has become routine enough to warrant a dedicated, sizable budget line rather than a one-off pilot. For an agency that faces essentially no public accounting for its purchases, "billions to test" is worth tracking even without the details filled in.
Until an actual number or vendor list leaks, billions is doing a lot of work in that headline. The real test will be whether any of this ever gets confirmed outside a classified estimate.