NSA AI testing costs billions as hospital coding adds $942 million

NSA AI-model testing is costing billions this year, while Blue Cross ties hospital AI coding to $942 million in extra spending over two years.

Sources disagree on how the NSA spending figure surfaced: one report says the agency told Congress it will spend billions this year testing advanced AI models, while another attributes that scale to sources familiar with classified estimates. Lawmakers fear broader federal AI oversight could reach tens of billions a year, far above earlier CBO scores of about $20 million a year for an AI risk center bill and $36 million over five years for a reporting system. Computing power and specialist hiring, including competition with AI lab pay, are the main costs, and the NSA Artificial Intelligence Security Center began frontier-model testing after high-profile hacking and security incidents. Separately, the Blue Cross Blue Shield Association ties hospital AI claims and coding tools to an additional $942 million over two years—also described as nearly $1 billion, or about $12,000 more per case—with a sharp rise in patients coded as complex and no evidence of a matching change in care. McLaren Health Care reports about $1 million more per month via SmarterDx, while insurers use AI to deny claims; Abridge founder Shiv Rao warned of agents fighting agents, and BCBSA’s Luke Chalker said insurers are on the losing side. OpenAI plans about $856 billion in compute through 2030, SemiAnalysis puts Anthropic’s API gross margin above 80 percent, and Google, OpenAI, and Anthropic are reportedly discussing a safety standards body that some experts reject as self-policing.

• Sources disagree on attribution: one account says the NSA told Congress it will spend billions this year testing advanced AI models; another cites sources familiar with classified estimates for the same scale of frontier-model evaluation.
• Lawmakers fear broader federal AI oversight could reach tens of billions annually, far above earlier CBO scores of about $20 million a year for an AI risk center bill and $36 million over five years for a reporting system.
• Compute and specialist hiring, including pay competition with AI labs, are the main costs; the NSA Artificial Intelligence Security Center began testing after high-profile hacking and security incidents.
• BCBSA links hospital AI coding and claims tools to $942 million in extra spending over two years, also described as nearly $1 billion and about $12,000 more per case, with more patients coded as complex and no matching change in care.
• McLaren Health Care reports about $1 million more per month via SmarterDx; insurers use AI to deny claims. Abridge founder Shiv Rao warned of agents fighting agents, and BCBSA’s Luke Chalker said insurers are on the losing side.
• OpenAI plans about $856 billion in compute through 2030, and SemiAnalysis estimates Anthropic’s API gross margin above 80 percent.
• Google, OpenAI, and Anthropic are reportedly discussing a safety standards body that some experts reject as self-policing.