Blue Cross links hospitals' AI coding tools to $942M in added healthcare costs
A claims analysis found no change in the actual care delivered, only in how it gets coded -- the share of inpatient stays billed as medically complex climbed from 37% to 40% as AI coding tools spread past 60% of hospital systems.
The Blue Cross Blue Shield Association published a claims analysis on September 25-26 estimating that hospitals' growing use of AI coding tools added $942 million in costs to its member plans over two years, with no corresponding change in the actual care patients received. The analysis found the share of inpatient stays billed to BCBSA plans as medically complex rose from 37% at the start of 2023 to 40% by the end of 2025, a period that coincides with more than 60% of hospital systems adopting AI tools that read lab reports and clinical notes to help assign billing codes.
Where the money actually goes
Of the $942 million, $653 million came from secondary diagnoses specifically -- codes that can hinge on a single lab value and are, per the analysis, particularly easy for an AI tool to surface and add, at roughly $11,000 per excess complex case. BCBSA's framing is blunt: a "clear disconnect between coding and treatment," where documentation intensity rose without any matching shift in what was clinically done for the patient. BCBSA senior vice president Luke Chalker called the dynamic a "completely one-sided blood bath" in disputes with hospitals over the disputed charges.
An AI vendor's response, and the shape of the fight ahead
Dr. Shiv Rao, founder of clinical-documentation AI company Abridge, acknowledged the dynamic could become "bots fighting bots, agents fighting agents" as insurers deploy their own AI to counter-audit hospital coding -- while also arguing that outcome could eventually lower costs rather than raise them. Whichever way that shakes out, this is a concrete, dollar-denominated version of a question that's mostly been asked abstractly until now: when AI gets good at optimizing one side of a billing dispute, the other side's AI spend on countering it is not hypothetical -- it's the next line item.