The AI Post
Agents & CodingOpen ModelsEnterpriseFundraisingGenerative MediaGovernanceInferenceInfrastructureLegal & SafetySector Impact
← Front Page Policy & Regulation · NSA · Encode

NSA tells lawmakers it spends billions testing AI models this year

The Washington Sun reports the figure from two people familiar with classified estimates, and says proposals for a civilian AI regulator are costed far lower.

The National Security Agency has told lawmakers it is spending billions of dollars this year to test and evaluate advanced AI models, The Washington Sun reported on Thursday, citing two people familiar with classified intelligence estimates. No other outlet has the figure, and the agency has not stated an amount publicly.

The comparison the piece draws is with the sums Congress has attached to civilian oversight. It puts the cost of the AI Security and Innovation Act at roughly $20 million a year, and a separate House bill on AI reporting at $36 million over five years. The reporting does not say how the NSA's classified spending is divided between models, staff and compute.

A Defense Department spokesperson declined to give a number. "For security reasons, the Department does not discuss the technical architecture or resource allocation for its AI tools," the spokesperson said, according to the report.

Nathan Calvin, general counsel at the advocacy group Encode, told the paper that in-house evaluation capability is "important and necessary" but "genuinely expensive". The report also quotes Nat Purser of the AI Verification and Evaluation Research Institute. Neither is in a position to confirm the classified figure.

The piece estimates that a full government AI regulatory system could run to tens of billions of dollars a year. That is the reporter's projection rather than a costing published by any agency, and no bill before Congress proposes a body at that scale.

Sources 1 source

  1. Source The Washington Sun