Blue Cross says hospitals' AI coding tools added $942M in health costs — with no sicker patients
A Blue Cross Blue Shield analysis of two years of hospital claims finds AI-assisted coding added $942 million in costs — driven by more complex bills, not sicker patients.
Artificial intelligence was supposed to make healthcare cheaper. According to the country’s largest health-insurance association, it has done the opposite — at least on the billing side. A new Blue Cross Blue Shield Association analysis says hospitals’ growing use of AI-assisted coding tools added roughly $942 million in costs to its member plans over two years — not because patients got sicker, but because their bills said they did.
The study, released Thursday, is the latest and sharpest volley in an escalating fight between insurers and hospitals over who is really winning the AI billing war. Hospitals say AI documentation tools help them capture the care they actually deliver. Blue Cross’s data suggests the tools are finding billable conditions — and charging more for them — without any matching change in the care patients received.
What the numbers show#
The association’s researchers analyzed de-identified inpatient claims from the first quarter of 2023 through the fourth quarter of 2025, comparing against a 2023 baseline. Their headline findings:
| Measure | What the study found |
|---|---|
| Total added cost | $942 million over two years vs. 2023 |
| From secondary diagnoses | $653 million — about 70% of the total |
| Excess complex cases | More than 55,000, averaging roughly $11,800 each |
| Share of stays billed “complex” | ~37% in early 2023 → ~40% by late 2025 |
| Partial intestinal blockage diagnoses | Up 55% among major bowel-surgery patients |
| Excess-acid diagnoses | Up 33% in the same group |
Anemia diagnoses also rose after major bowel procedures — without any corresponding rise in blood transfusions, the standard treatment for the condition. That mismatch is the thread the researchers keep pulling: if patients were genuinely sicker, more treatment should show up somewhere in the data. It doesn’t.
How AI got into the coding loop#
Two kinds of AI are doing the work, according to the association. Ambient scribes passively listen to conversations between patients and clinicians and draft medical notes. Other tools scan existing patient records and flag secondary conditions — illnesses that coexist with the main reason for admission — that a clinician might not have documented.
Why does that matter for money? In hospital billing, secondary conditions can push a stay into a higher-severity diagnosis-related group, the category that determines reimbursement. A more complex bill commands a higher payment. AI is unusually good at this particular job: it can surface a condition from structured, objective data — say, a low hemoglobin reading that suggests anemia — without needing a clinician to judge symptoms first. Billing for a real condition a patient actually has is standard practice. The question the study raises is whether every flagged condition is one the patient was actually being treated for.
The diagnoses without the treatment#
“If patients are truly sicker, we’d expect to see more treatment,” said Luke Chalker, the association’s senior vice president of product and data science, one of the analysis’s authors. “The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.”
This is the association’s second round of research on the theme. In March, it published an analysis focused on posthemorrhagic anemia in maternity cases, where diagnoses surged at a subset of hospitals without a corresponding rise in treatment. That work projected roughly $663 million in excess inpatient spending nationally — and flagged at least $1.67 billion of outpatient exposure. The new study widens the lens from one diagnosis to overall coding intensity, and lands on the same conclusion: “Critically, what we found is underneath all of that data [was] no change in corresponding care for a more complex patient,” Chalker told reporters. The reimbursement machinery, he said, is what allows the gap to persist.
An insurer–hospital billing war#
The study landed in the middle of what hospital-trade reporting calls an “AI billing war.” Hospitals have argued they are the ones losing it — that AI is mostly a weapon insurers use to deny claims and downcode stays. Blue Cross is now presenting the mirror image: AI-assisted coding on the provider side, pushing bills up.
The caveat is worth stating plainly: the Blue Cross Blue Shield Association pays these bills. Its 31 member companies cover more than 100 million people, so it has both the data and the financial motive to scrutinize coding intensity. But it is not alone in the complaint — insurers including Centene have told investors that health systems’ use of AI tools has produced aggressive or inappropriate reimbursement payments. And Chalker argues the consequence lands on someone other than the combatants: unnecessary spending turns into higher premiums and out-of-pocket costs for enrollees, employers, and taxpayers.
What to watch#
Three questions follow the $942 million figure. First, the outpatient side: the March analysis’s $1.67 billion outpatient estimate suggests inpatient coding is only the visible part of the phenomenon. Second, the hospital response — whether providers dispute the methodology or adjust their own AI coding practices. And third, whether regulators or federal payers decide that AI-assisted coding intensity is something to audit rather than admire.
The bigger shift is harder to measure than any dollar figure. Ambient scribes and coding assistants were sold to the industry as efficiency technology — tools to reduce paperwork and clinician burnout. Blue Cross’s analysis is a reminder that in American healthcare, the tool that writes the note also shapes the bill, and the line between capturing real complexity and manufacturing it may be getting blurrier by the quarter.