Insurers Say Hospital AI Coding Added $942 Million in Costs — Hospitals Say Their Patients Really Are Sicker
The first large-scale audit of AI in American medical billing has landed, and it reads like a preview of what happens when two industries point algorithms at each other’s paperwork. The Blue Cross Blue Shield Association — the federation whose member plans cover about one in three Americans — released an analysis on September 24, 2026 estimating that hospital coding practices added roughly $942 million in extra healthcare spending across its plans between 2023 and 2025. Its claim, per the BCBSA’s own release: “coding for care has changed, but there is no evidence of corresponding change in care delivered.”
The insurer group ties the shift directly to AI. More than 60% of US hospital systems now use AI coding tools — software that scans lab results, doctors’ notes and electronic records to find secondary diagnoses — and TechCrunch reports the analysis found a sharp rise in patients being documented as complex, without matching treatment. The flagship example from the analysis: significantly more anemia diagnoses at hospitals performing major bowel surgery, “without a corresponding increase in transfusions.” About 70% of the added cost — roughly $650 million — came from secondary diagnoses that pushed claims into higher-reimbursement categories, per the BCBSA’s breakdown of its claims data.
The Other Side of the Argument
The hospital industry does not accept the framing. As Medical Daily summarises the dispute, hospital representatives argue today’s inpatients are genuinely older and sicker, and that better documentation simply captures conditions that were missed before — the 37%-to-40% rise in complex-coded inpatient claims reflects clinical reality, not software. Neither side has released patient-level chart reviews that would settle the question, which is worth holding onto: this is an insurer’s analysis of insurer’s claims data, contested by the people being audited. Luke Chalker, BCBSA’s senior vice president of product and data science, made the group’s position plainly: “The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.” The hospital industry’s counter is that the disconnect has other causes — an aging population, rising chronic disease, milder cases shifting outpatient.
Bots Fighting Bots
The most quoted line of the week belongs to Dr Shiv Rao, founder of medical-AI startup Abridge, who told TechCrunch the trajectory could produce “a horrible dystopic future nobody wants to live in” — “bots fighting bots, agents fighting agents.” His caveat: the same dynamic might eventually reduce tension and cut costs. BCBSA’s Chalker, meanwhile, rejected even the war metaphor in the opposite direction: “It’s not a war. It’s a completely one-sided blood bath” — with insurers, the party running the analysis, cast as the losing side.
That framing deserves scepticism in both directions. Insurers are not neutral observers of billing intensity — they pay the claims — and they run aggressive AI of their own on the other side of the transaction, including automated review of what care to authorise. When both parties to a negotiation automate aggressively, each system optimises against the other, and the bill — premiums, out-of-pocket costs, taxpayer-funded programmes — lands on everyone else. The BCBSA analysis is best read as one combatant’s evidence brief in a fight where both sides are armed with the same class of tools.
Why It Matters Here
New Zealand’s health system runs on completely different financing — no DRG upcoding arms race of this shape — but it is actively importing the same tooling. Health NZ’s HealthX programme has run AI scribe trials in emergency departments that cut after-hours admin by up to 81% and let clinicians see an extra patient per shift, and the scribe rollout has since expanded nationally after pilots in Hawke’s Bay and Whanganui. Ambient scribes are precisely the technology class the BCBSA flags as able to surface secondary diagnoses “through listening during visits” — which is a benefit or a billing risk depending entirely on what the billing rules reward. The American lesson is not “don’t use scribes”; it is that the same AI that saves a clinician an hour can also, silently, reclassify the patient, and the audit system has to be built for that before the volume arrives, not after.
The second lesson is structural. As we noted in our coverage of AI labs quietly absorbing tens of thousands of incidents, the pattern across 2026 is AI harms surfacing in administrative plumbing long before anything cinematic happens. The BCBSA figures are the billing-form version of that: not a rogue agent, not a jailbreak — just optimisation pressure, applied relentlessly at the seam where software meets reimbursement. Regulators in the US are now looking at whether AI-assisted coding warrants audit requirements; the EU’s AI Act transparency rules already push in that direction. New Zealand, with no equivalent disclosure regime yet, is watching this dispute from inside the experiment — its own deepfake and AI transparency debates are early chapters of the same problem: systems that make claims about reality need to be auditable by someone who is not a party to the transaction.
❓ FAQ
What did the BCBSA analysis actually find?
That the share of inpatient claims coded as medically complex rose from about 37% at the start of 2023 to about 40% by the end of 2025 across Blue plans, adding an estimated $942 million in costs over two years against a 2023 baseline — with roughly $650 million of that tied to secondary diagnoses, and no corresponding increase in treatments delivered.
Does the analysis prove AI caused the increase?
No. It shows rising coding intensity correlating with AI coding tool adoption — more than 60% of hospital systems now use such tools — and points to examples like rising anemia codes without rising transfusions. The American Hospital Association argues aging patients, chronic disease and outpatient shifts explain the change. No patient-level chart review has been released by either side.
What is the NZ connection?
Health NZ already uses AI scribes nationally in emergency departments after successful trials. The US dispute shows the downstream question — what AI-surfaced diagnoses do to billing and records integrity — is one health systems everywhere will face as clinical AI documentation becomes standard.
Sources: BCBSA analysis (September 24, 2026), TechCrunch, Medical Daily, Reuters (paywalled), Fierce Healthcare (paywalled). The BCBSA figures are the insurer group’s own claims-data analysis; the hospital industry disputes the interpretation and neither side has released patient-level chart reviews.
