AI raises costs at NSA, hospitals, and insurance companies

By Splitfeed EditorialLast updated Sep 26, 2026, 19:06 UTC

Around September 25, 2025, reporting attributed to The Washington Sun revealed that the NSA is already spending billions of dollars to test advanced AI models, with the bulk of those costs tied to computing infrastructure rather than software or personnel. Separately, AI-assisted billing code tools in the US healthcare sector drove up costs by nearly $1 billion over two years. On the legislative side, lawmakers now project that full-scale federal AI oversight will cost tens of billions of dollars annually, a figure that dwarfs the Congressional Budget Office's earlier estimate of just $20 million.

The gap between the CBO's $20 million figure and the tens-of-billions projection lawmakers are now working with is the central dispute here. That gap is not a minor rounding error; it represents a fundamental disagreement about the scope of what AI oversight actually requires at federal scale. The healthcare cost increase adds a separate but related dimension, showing that AI adoption in regulated industries can quietly inflate expenditures even when no single line item looks alarming. Together, these data points raise a genuine question about whether early cost modeling for AI adoption in government and healthcare was structurally flawed or simply based on assumptions that have since collapsed.

Because no reactions from any camp had been published at the time of this report, what follows reflects what each camp would typically argue about a story like this. Pro-AI advocates would likely frame the cost increases as growing pains, arguing that computing investment at the NSA reflects necessary modernization and that efficiency gains will eventually offset early expenditures. Anti-AI critics would almost certainly treat the numbers as confirmation that AI deployment is being rushed without adequate fiscal accountability, pointing to the billing code cost increases as evidence of real harm already accruing. The middle ground camp would probably call for better cost-benefit analysis and independent auditing before further expansion, rather than either accelerating or halting adoption.

The figure to watch is whatever the CBO or a relevant congressional committee produces as a revised cost estimate for federal AI oversight, and whether the healthcare billing data prompts any regulatory response from CMS or similar bodies. A formal budget reconciliation for AI oversight spending, if one is scheduled, would go a long way toward settling the dispute over what full-scale adoption actually costs.