The news
The Blue Cross Blue Shield Association (BCBSA) said on September 24 that, by its estimate, a jump in patients billed as medically complex added about $942 million in spending for Blue plans from 2023 to 2025, a rise it linked to hospitals' growing use of AI coding tools.
The analysis, built from de-identified claims data from Blue Cross and Blue Shield companies that cover one in three Americans, found that coding shifted while care stayed largely the same. BCBSA said it saw no sign of a matching change in treatment and called the gap a clear disconnect. The release does not show which individual claims involved AI tools.
Most of the increase came from secondary diagnoses, conditions recorded in addition to the main reason for a hospital stay, which can move a claim into a more severe, better-paid billing category. BCBSA said about 70% of the added cost, or more than $650 million, was tied to such diagnoses. Some can be derived from a single lab value, which the association said makes them well suited to detection by AI software that scans lab results, doctors' notes and electronic records.
The group said more than 60% of hospital systems have started using AI coding tools. Its example: significantly more anemia diagnoses after major bowel surgery without a matching rise in blood transfusions. Luke Chalker, BCBSA's senior vice president of product and data science, said the gap suggests "AI is identifying more billable conditions, not sicker patients." David Merritt, BCBSA's senior vice president of external affairs, said the findings show the need to study these tools more closely and whether they are worsening healthcare affordability.
The hospital industry disputes the premise. In a July 31 fact sheet published before this analysis, the American Hospital Association (AHA) called insurer claims that AI tools have driven up coding intensity unsubstantiated. It attributed more complex hospital coding to an aging population, more chronic disease and routine cases moving to outpatient care.
The report builds on earlier BCBSA research and extends it to major bowel procedures. TechCrunch, covering the findings on September 26, noted that hospitals and insurers are both using AI in billing disputes, and quoted Abridge founder Shiv Rao as warning of a future of bots fighting bots while arguing AI might also reduce tensions and cut costs.
The numbers
- Estimated added spending, 2023-2025
- $942 million
- Share tied to secondary diagnoses
- About 70% (more than $650 million)
- Hospital systems using AI coding tools
- More than 60%
- Americans covered by Blue plans in the data
- One in three
Why CEOs should care
For CFOs and benefits leaders at self-insured employers, this is an AI cost that lands on your health plan, not your IT budget. Ask your insurer or claims administrator whether they track coding intensity by hospital, especially secondary diagnoses that raise payment levels, and whether they audit those claims against treatment records. Build the question into your next plan renewal, and ask for the data behind any year-over-year change in severity mix. Employers that negotiate directly with hospital systems should ask the same questions of their providers.
Hospital and health system executives face the mirror-image risk. AI documentation tools can capture real conditions that were previously missed, but payers are now publishing evidence that coding rose without matching treatment. Chief financial and compliance officers should require vendors to show an audit trail linking each added diagnosis to clinical evidence, monitor coding changes against care patterns, and prepare for tougher claim reviews.
For boards and technology buyers in any industry, the broader lesson is that AI deployed in an adversarial process such as billing, procurement or claims can shift costs to the other side rather than create value. Before scaling AI in revenue-facing workflows, ask how counterparties are likely to respond, and whether the gains will hold up once they deploy their own tools.
The bigger picture
Healthcare is an early test of what happens when both sides of a transaction automate. Hospitals use AI to document and code claims, and TechCrunch, citing The New York Times, reported that AI use by both hospitals and insurers appears to be worsening long-running payment fights. The same dynamic could appear in other industries where AI is used to maximize one party's claim, from insurance adjusting to vendor invoicing, and it will shape how regulators and customers judge the productivity case for AI.
What’s next
Expect insurers to use findings like these to push for payment-policy changes and more audits of secondary diagnoses. Watch whether hospital groups or AI coding vendors answer this analysis with their own data. Employers renewing plans for 2027 should ask whether their carriers are adjusting for coding intensity and how that shows up in premiums.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
