Washington Sun report: NSA spending billions annually on classified AI cyber testing
A September 24 Washington Sun report — which continued to draw attention on September 26 — says the NSA is spending billions annually on classified frontier AI cyber testing; the exact figure is not verified by public budget documents.
On September 24, the Washington Sun reported, citing two people familiar with classified intelligence estimates, that the NSA will spend billions of dollars this year testing frontier AI models for cyber risks. The story continued to spread on September 26 via outlets such as The Automated Daily, raising Capitol Hill concerns about the long-term cost of any federal AI oversight regime.
The testing work sits inside NSA’s Artificial Intelligence Security Center (AISC) and targets cyber capabilities vendors disclose but do not independently verify. Executive Order 14409 directs Treasury acting through the NSA director and DHS acting through CISA to establish and maintain a classified cyber-capability benchmarking process, but the order also makes clear the framework is voluntary and does not constitute mandatory licensing, pre-clearance, or permitting.
Ground Truth’s September 26 independent check noted the "billions" figure rests on only two unnamed sources and is not corroborated by public budget documents, contracts, vendors, models, or specific amounts. The same review pointed out that the Congressional Budget Office’s estimate for a separate AI risk center stood at around $20 million per year — a wide gap that leaves the actual scale and composition of any federal AI oversight regime still opaque.