Sources: the NSA told lawmakers it is spending billions this year to test AI models; proposals for a US AI regulatory body estimated costs of $20M-$40M per year
The price tag is significantly higher than previously known. — Copy — The National Security Agency told lawmakers …
Context & Ripple Effects
The NSA had already framed protection of American AI models as a national-security issue when it announced an AI security center in 2023. A 2024 bipartisan Senate plan also put federal AI capacity—including major R&D spending and a privacy law—on the policy agenda.
The reported testing budget puts a concrete, classified procurement dimension behind that agenda. It also follows US AI labs' warnings that unauthorized model distillation is a major threat, while the proposed regulator's $20 million–$40 million annual cost remains unconfirmed.
First-order effects
- According to the sources, the NSA becomes a multibillion-dollar buyer of AI-model testing capacity in 2026, concentrating government demand around evaluation rather than only AI policy design.
- Lawmakers considering a US AI regulator must weigh its reported tens-of-millions operating estimate against an intelligence agency's far larger, classified model-testing spend; the regulator estimate is unconfirmed.
Second-order effects
- AI labs that have warned the White House about unauthorized model distillation costs gain a clearer federal-security rationale for testing and protection work, alongside commercial model development.
- The NSA's spending makes evaluation compute, model access, and security testing more consequential procurement categories for suppliers seeking US government AI business.
Third-order effects
- If intelligence-agency testing and a civilian regulator develop in parallel, US AI governance may split between classified national-security evaluation and public-facing oversight rather than residing in a single body.
- Federal AI policy is moving from broad ambitions in the bipartisan Senate AI plan toward sustained state purchasing of the infrastructure and assurance work needed to deploy models.
The trend: US AI policy is becoming state-mediated AI industrial policy, with security evaluation and procurement gaining weight alongside formal regulation.