Workers’ compensation carriers and self-insured employers pay hospital bills under fee schedules built on the same diagnosis codes that group health insurers use, so a dispute now playing out between the nation’s largest insurer federation and the hospital industry deserves close attention from claims and bill review professionals. On September 24, 2026, the Blue Cross Blue Shield Association (BCBSA) released a claims analysis concluding that hospitals’ growing use of artificial intelligence to document and code inpatient stays has made patients look sicker on paper without any matching change in the care they received. As Fierce Healthcare reported, the association put the price of that shift at an estimated $942 million in added costs to Blue plans over 2024 and 2025, measured against a 2023 baseline.
To follow the argument, it helps to understand how a hospital stay is priced. Most inpatient admissions are paid as a single lump sum determined by a diagnosis-related group, or DRG. Each DRG family typically comes in tiers. The same operation pays one amount for an uncomplicated patient, more if the record lists a “complication or comorbidity,” and more still if it lists a “major” one. A single secondary diagnosis, such as anemia from blood loss, a low sodium level, or malnutrition, can move a stay up a tier and add thousands of dollars to the payment even though the surgery itself is unchanged.
That is where the new software comes in. AI revenue cycle tools scan physician notes, laboratory results and other chart entries for conditions that qualify as secondary diagnoses, and ambient “scribe” products listen to patient encounters and draft the clinical note. Hospitals say these tools capture conditions that busy clinicians always treated but did not always write down. Insurers say they are finding billable diagnoses that made no difference to the patient’s care.
BCBSA’s analysis, a short white paper covering claims from the first quarter of 2023 through the end of 2025, found that the share of inpatient cases billed to Blue plans as medically complex rose from 37 percent to 40 percent. About 70 percent of the increase came from more than 55,000 additional cases in which a secondary diagnosis pushed the claim into a higher-paying DRG. Those cases accounted for $653 million of the total, or roughly $11,000 per case. The paper used major bowel surgery as its example: claims in the highest severity tier climbed to 22.7 percent while uncomplicated cases fell from 36.6 percent to 32.8 percent.
The association’s central evidence is what it calls a disconnect between coding and treatment. Hospitals in the top quarter for growth in complex coding showed similar or lower rates of intensive care use, transfusion, reoperation and length of stay than their peers. For acute blood loss anemia, which BCBSA describes as a common “bump” code, the hospitals that diagnosed it most often actually transfused those patients less often, 16.9 percent compared with 19.3 percent elsewhere. Luke Chalker, the association’s senior vice president of product and data science, told reporters the data showed “no change in corresponding care for a more complex patient.” The September paper follows a March 2026 BCBSA and Blue Health Intelligence study that attributed about one-fifth of a 9 percent rise in per-member inpatient costs to coding intensity.
Hospitals reject the conclusion. In an October 5 response, the American Hospital Association said patients today are older and more clinically complex and that AI tools help providers record their conditions accurately. “The question is whether the patient’s medical record supports those diagnoses,” two AHA policy executives wrote, noting that BCBSA reviewed claims rather than charts and did not identify which claims actually involved AI. The AHA made similar points in an August fact sheet, and in a statement to CNBC it criticized insurers for raising coding concerns while relying on their own automated downcoding and denial systems.
The study’s limits are real, and BCBSA concedes the main one. It relied on claims data, not medical records, so it cannot show that any particular diagnosis was unsupported. The link to AI rests on timing and on survey figures showing that most hospital systems now use AI somewhere in the revenue cycle, not on claims traced to specific software. The white paper is not peer reviewed, it names no hospitals, and it arrived during a period of difficult contract negotiations between plans and hospital systems. Independent researchers had nonetheless raised the same concern before the insurers did. A January 2026 viewpoint in JAMA Health Forum warned that ambient scribes marketed for their ability to capture more and more severe diagnoses could drive spending upward, and an April report from the Peterson Health Technology Institute described a payer-provider AI “arms race” that raises billing intensity without fixing underlying inefficiency.
For workers’ compensation payers the relevance is direct, even though the study examined commercial health claims and its examples were bowel surgery and, earlier, maternity care. California’s Official Medical Fee Schedule prices inpatient hospital stays using Medicare’s DRG system with a multiplier, as do the fee schedules of many other states. The DRG families that dominate comp inpatient spending, including spinal fusion, major joint replacement and fracture repair, are tiered by complications and comorbidities in the same way. Acute blood loss anemia, the code BCBSA singled out, is one of the diagnoses most frequently recorded after orthopedic surgery. A hospital that deploys AI coding software applies it to every payer’s claims, not just those of Blue plans.
That raises several practical points. First, bill review programs that validate only the fee schedule calculation will not catch severity drift, because a correctly priced DRG can still rest on a secondary diagnosis the record does not support. Clinical validation, which compares the coded diagnoses against the treatment actually delivered, is the test BCBSA applied in aggregate and the one payers can apply claim by claim. Second, any objection must still be made within the deadlines and explanation-of-review requirements of Labor Code § 4603.2, and unresolved payment disputes proceed through second review and independent bill review. Third, the issue cuts both ways. Applicants’ attorneys and providers will point out that more complete documentation of comorbidities is legitimate and may bear on apportionment, causation and future medical needs, while defense counsel may question whether a newly charted secondary condition was ever clinically significant.
Payers should also expect scrutiny of their own automation. According to an Orthopedics This Week commentary, Indiana now bars health plans from using an automated tool as the sole basis for downcoding a claim without reviewing the medical record, and lawmakers in California and several other states introduced downcoding bills this year. Those measures are aimed at health plans, but they signal where regulators are heading.
BCBSA says further analyses are coming, including outpatient care and additional DRGs, and orthopedic and spine procedures are obvious candidates. The broader cost pressure is not in dispute. CNBC reported that benefits consultant Marsh projects employer health costs per employee will rise 8.2 percent in 2027. Whether AI-assisted coding represents accurate documentation finally catching up with reality or technology-enabled upcoding is a question the claims data alone cannot answer. For comp payers, the prudent step is to find out whether their bill review vendor tracks DRG severity mix over time and tests secondary diagnoses against the treatment record.