AI upcoding now has a price tag, according to the insurers that cover one in three Americans. On 24 September 2026 the Blue Cross Blue Shield Association (BCBSA) published an analysis estimating that hospitals billing inpatient stays as more medically complex added about $942 million to what Blue Cross and Blue Shield companies paid over two years. The association links that rise to AI tools that scan patient records and lab results for anything that can be coded.
The core claim is a mismatch. Between early 2023 and late 2025, the share of Blue members’ inpatient claims billed as complex rose from roughly 37% to about 40%, yet BCBSA says the treatment patients received did not change to match. Around 70% of the extra cost came from secondary diagnoses, often derived from a single lab value, that moved a claim into a higher-paying category. Hospital groups reject the AI upcoding charge outright, arguing that their patients are older and sicker and that insurers are using the claim to cut payments.
This article sets out what the new white paper measured, how AI upcoding would work on an individual claim, and the clinical tests BCBSA used to back its case. It also explains why some headlines say $2.3 billion, what the American Hospital Association says in reply, and who ultimately pays.
Table of contents
- What the Blue Cross AI Upcoding Report Found
- How AI Upcoding Works: DRGs and Bump Codes
- Major Bowel Procedures: The AI Upcoding Case Study
- The Clinical Discordance Test Behind the AI Upcoding Claim
- $942 Million or $2.3 Billion? Two AI Upcoding Reports
- The Hospital Case Against the AI Upcoding Charge
- Who Pays for AI Upcoding: Premiums, Employers and Taxpayers
- What Comes Next for AI Upcoding Oversight
- AI Upcoding FAQ
- References
What the Blue Cross AI Upcoding Report Found
The document behind the AI upcoding headlines is a three-page white paper titled Hospital Coding Intensity Analysis: Major Bowel Procedures, dated September 2026. Its authors are Chris Birkmeyer, David Wennberg, Keith Kamons and Luke Chalker. It covers inpatient claims from the first quarter of 2023 to the fourth quarter of 2025, drawn from Blue plans that together cover about one in three Americans. BCBSA released it with a press statement and a briefing for reporters.
$942 million over two years
The headline number is an estimate of incremental claims costs to the Blue System alone, not to the US health system as a whole. It is measured against a 2023 baseline, so it captures what hospitals were paid in 2024 and 2025 above what the 2023 coding mix would have produced. Reuters framed it as “nearly $1 billion” in extra health spending, which is where most headlines came from.
Luke Chalker, a BCBSA senior vice president and one of the white paper’s authors, told reporters the estimate “is purely for [where] we believe there is no change in care delivered,” according to Fierce Healthcare. Anything that led a hospital to document additional care was left out. “That’s the stuff that hospitals should bill for, and that’s the stuff we should pay for,” he said.
55,158 extra complex cases
The most specific number in the white paper is a count. Compared with 2023 rates, hospitals classified 55,158 additional cases as complex, meaning they carried a complication or comorbidity (CC) or a major complication or comorbidity (MCC). Those cases generated $653 million in incremental reimbursement, an average of $11,800 per excess complex case. Dividing $653 million by 55,158 gives roughly $11,840, so the arithmetic holds.
That $653 million is the part BCBSA ties to AI upcoding most directly, because it comes from secondary diagnoses that bump a claim into a higher-severity payment group. It works out to 69% of the $942 million total, which the association rounds to 70%. The three-page white paper does not break down the remaining $289 million or so.
Why BCBSA says care did not change
The argument rests on what the association calls clinical discordance: a diagnosis appears on the claim, but the treatment you would expect for it does not. “If patients are truly sicker, we’d expect to see more treatment,” Chalker said in the press release. “The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.”
BCBSA also points to adoption. It says more than 60% of hospital systems now use AI-enabled tools that can scan lab results and electronic records to find secondary diagnoses. The timing of that adoption, and the type of codes that grew fastest, are the two pillars of the AI upcoding case.
How AI Upcoding Works: DRGs and Bump Codes
To see why one extra diagnosis matters, you have to look at how US hospitals are paid for inpatient care. Most stays are reimbursed through a diagnosis-related group (DRG), a fixed payment category assigned to the whole admission. BCBSA’s research notes that hospitals are primarily reimbursed for inpatient care this way, including by commercial insurers.
Three tiers, three prices
Many DRG families come in three severity tiers: a base version, a version “with CC” and a version “with MCC”. The Medicare MS-DRG system maintained by CMS defines which secondary conditions count. Adding a qualifying secondary diagnosis can move an admission up a tier, and the higher tier pays more because it assumes the patient needed more resources. That is the lever the AI upcoding debate turns on.
The bump codes in the white paper
BCBSA calls the diagnoses that trigger a tier change “bump codes”. For major bowel procedures, the fastest-growing ones included several that can be read straight off a lab result, alongside others that depend on written documentation.
| Secondary diagnosis | ICD-10 code cited | Source type named by BCBSA |
|---|---|---|
| Unspecified acidosis | E87.20 | Lab-derived |
| Hypo-osmolality and hyponatremia (low sodium) | E87.1 | Lab-derived |
| Acute posthemorrhagic anemia | D62 | Lab-derived |
| Malnutrition | Not cited | Documentation |
| Abdominal abscess | Not cited | Documentation |
| Intestinal obstruction | Not cited | Documentation |
Why lab-derived diagnoses suit AI tools
“Many of these diagnoses can be derived from single laboratory values or routine observations, making them particularly well-suited for detection by modern RCM technology,” the white paper says. RCM stands for revenue cycle management, the back-office work of turning care into a bill. BCBSA singles out two technologies: laboratory data mining, and ambient listening tools that use speech recognition and natural language processing to turn a clinician’s conversation with a patient into structured notes in the electronic health records.
Coding software can then scan those notes and lab results and propose extra billable diagnoses, the workflow described in our guide to AI medical coding software. The concern is not that the software invents conditions. A lab value outside the normal range is real. The question BCBSA raises is whether a mild abnormality that no clinician treated should change the price of the stay.
The white paper is specific about this. When AI is used to identify abnormal tests, “it may be set to flag all tests that have results outside of the normal range, even though clinicians typically do not treat mild anemias.” In the AI upcoding argument, that gap between a flagged value and a treated condition is everything.
Major Bowel Procedures: The AI Upcoding Case Study
BCBSA chose one DRG family to show the AI upcoding pattern in detail. Major bowel procedures, DRGs 329 to 331, are surgeries used mainly to treat colon cancer and diverticular disease. Because the family has three severity tiers, it shows clearly how claims move between them over time.
The top tier grew as the bottom tier shrank
Between 2023 and 2025, claims at the highest level of complexity (with MCC) rose from 20.2% to 22.7% of the family, while non-complex cases fell from 36.6% to 32.8%. The white paper says this happened “even though the associated treatments themselves did not change.” Some early coverage gave the starting MCC figure as 10.2%. The white paper itself says 20.2%, which makes the rise 2.5 percentage points rather than 12.5.
The case mix moved upward at both ends: more claims in the top tier, fewer in the bottom one.
$60.8 million from a single surgical family
This one DRG family accounts for $60.8 million of incremental claims costs in BCBSA’s estimate, as patients moved into higher-severity, higher-reimbursement categories. That is about 6.5% of the $942 million total, from a single surgical category. BCBSA says the pattern matches broader trends across inpatient claims, and Chalker said the association plans further analyses of other DRGs and of outpatient care.
The Clinical Discordance Test Behind the AI Upcoding Claim
A rise in complex coding proves nothing about AI upcoding on its own. Patients could be getting sicker, or hospitals could be recording conditions they previously missed. BCBSA’s answer is to compare what was coded with what was done, hospital by hospital.
Top-growth hospitals versus their peers
The white paper splits hospitals into the 25% with the fastest complexity growth and the rest, then compares their 2025 bowel procedure claims. If the top group’s patients were truly sicker, you would expect more intensive care, more transfusions and longer stays. The data shows the opposite, or no difference at all.
| 2025 bowel procedures | Top 25% hospitals | Other hospitals | Gap |
|---|---|---|---|
| Coded complex (MCC or CC) | 75.6% | 65.0% | +10.6 points |
| ICU use | 11.5% | 13.2% | -1.7 points |
| Transfusion | 3.6% | 3.9% | -0.3 points |
| Reoperation | 1.7% | 1.5% | +0.2 points |
| Median length of stay | 4.0 days | 4.0 days | None |
| Anemia diagnosis rate | 13.7% | 9.9% | +3.8 points |
| Transfused, when anemia was coded | 16.9% | 19.3% | -2.4 points |
This table is the core of BCBSA’s AI upcoding evidence, and the white paper draws its strongest conclusion from it. “The consistent inverse relationship between diagnosis-based complexity and both aggregate resource utilization and diagnosis-specific procedural intervention is the strongest indicator that coding escalation reflects documentation practice changes rather than actual patient acuity shifts,” it says.
The anemia signal
Posthemorrhagic anemia, low haemoglobin after blood loss, often occurs during major surgery, and BCBSA calls it “a common bump code”. Top-growth hospitals coded it in 13.7% of bowel procedure cases against 9.9% elsewhere, a rate 38% higher. Yet among patients given that diagnosis, fewer were transfused at the top-growth hospitals: 16.9% compared with 19.3%.
More anemia appeared on the bills, and less of it was treated with blood.
Razia Hashmi, BCBSA’s vice president of clinical affairs, framed the question at the briefing. “The question that is worth asking is [with] two similarly situated hospitals, treating similar patients, why would one hospital diverge?” she said, according to Fierce Healthcare. “There may be an element of correct coding there, but the likelihood that this is technology-enabled upcoding is higher, in my view.”
What the claims data cannot show
BCBSA acknowledges the main limitation. The analysis uses claims, not clinical records, so it cannot check whether each patient was sicker than the coding suggests. Chalker said chart review would strengthen the case, and that Blue plans with access to clinical data have “been able to kind of re-emphasize and demonstrate this effect.”
The September paper also does not identify which hospitals use which tools, or name any vendor. The link to AI upcoding rests on timing, the kind of codes involved and adoption surveys, not on claims traced one by one to a piece of software. That distinction is exactly where the hospital industry has aimed its response to the AI upcoding charge.
$942 Million or $2.3 Billion? Two AI Upcoding Reports
Readers following this story may have seen a much larger figure. Some coverage on 24 September reported that AI coding tools drove about $2.3 billion in excess costs across 62 million members. That number is real, but it comes from an earlier BCBSA study published in March 2026, not from the new white paper.
The March maternity study
The March issue brief, Rising Coding Intensity and Its Impact on Health Care Affordability, was produced by BCBSA’s analytics partner Blue Health Intelligence (BHI). It analysed commercial inpatient claims from April 2022 to March 2025 across plans covering about 62 million members, and used maternity care as its case study.
- The 10% of hospitals with the biggest jump in complex coding went from 46.8% of admissions coded as complex in the second quarter of 2022 to 59.8% by the first quarter of 2025. The other 90% rose from 51.2% to 55.4%.
- At hospitals with the fastest growth in postpartum anemia coding, the share of maternity admissions carrying the diagnosis rose from 4.0% to 12.3%, while transfusions moved only from 0.8% to 1.2%.
- One Blue plan audited an outlier hospital system where maternity anemia coding rose from 2.9% to 23.5% of admissions. Working with a board-certified OB-GYN, it found that fewer than 20% of the audited cases met established clinical criteria.
Why the headlines disagree
The two reports measure different things. The March brief estimated that about $663 million in inpatient spending and at least $1.67 billion in outpatient spending “may be tied to more aggressive, AI-enabled coding practices nationwide”, which together make roughly $2.3 billion. The September white paper is narrower: $942 million in inpatient costs, to the Blue System only, over 2024 and 2025. Blending the two produces headlines that are each partly right and hard to compare.
| Feature | March 2026 issue brief | September 2026 white paper |
|---|---|---|
| Produced by | Blue Health Intelligence for BCBSA | BCBSA (Birkmeyer, Wennberg, Kamons, Chalker) |
| Claims period | April 2022 to March 2025 | Q1 2023 to Q4 2025 |
| Case study | Maternity admissions, postpartum anemia (D62) | Major bowel procedures (DRGs 329 to 331) |
| Hospital cohort compared | Top 10% by complexity growth | Top 25% by complexity growth |
| Headline cost | About $663M inpatient plus at least $1.67B outpatient, nationwide | $942M inpatient, Blue System, two years |
| Case-study cost | $22M (maternity anemia) | $60.8M (bowel procedure family) |
The estimates range from $22 million to $1.67 billion depending on what is being counted, so the scope matters more than any single number.
How the March figures fit the new ones
The March brief also estimated that per-member inpatient costs rose about 9% from 2023 to 2024 across the participating plans, and that about a fifth of that increase came from rising coding intensity. That works out to roughly 1.8 percentage points of the 9%. BCBSA’s summary of the study cited one facility whose complexity rating rose 6.7% after it announced a switch to AI, against 0.9% at other facilities in the same state. The September paper, BCBSA says, “finds consistent results” in a different surgical category, which is why the association now presents AI upcoding as a widespread pattern rather than a maternity quirk.
The Hospital Case Against the AI Upcoding Charge
Hospitals were pushing back against the AI upcoding narrative before this report landed. In a fact sheet on AI and coding intensity dated August 2026, the American Hospital Association (AHA) called such claims “unsubstantiated”. On 8 September, Inside Health Policy reported that the AHA had asked Congress and the Trump administration to stop insurers from falsely accusing providers of AI upcoding.
Sicker patients, not smarter software
The AHA argues that coding intensity is rising because patient acuity is rising. It cites an ageing population, more chronic disease and the continued shift of simpler care to outpatient settings, which leaves hospitals with sicker inpatients. According to the fact sheet, AHA data show 19% of hospital expense growth from 2019 to 2024 reflects caring for sicker, more complex patients. An AHA and Vizient analysis found the hospital case-mix index rose about 5% over the same period.
The AHA also points to changes in coding guidelines that encourage more specific, higher-acuity diagnoses. “Hospitals are increasingly caring for patients with higher acuity, a trend that is appropriately reflected in provider coding practices,” it wrote.
Downcoding and the claims arms race
The hospital group’s sharper point concerns insurers’ own conduct. It says some commercial plans run “downcoding programs” that use automated edits to cut payment for higher-level care without reviewing medical records, forcing providers into costly appeals. Hospital leaders have described AI coding tools partly as a defence against denials and payment delays. Fierce Healthcare described the result as an AI arms race over claims, with payers and providers each deploying software against the other.
The insurers’ own coding record
The AHA also turns the charge around. It notes that insurers have told investors their own enrollees are sicker when it comes to risk scores. “In short, these plans want it both ways — a sicker enrollee population for purposes of health plan risk scores but a healthier one when it comes time to cover healthcare claims,” the fact sheet says.
It cites a 2025 MedPAC finding that upcoding contributed to $40 billion in excess payments to Medicare Advantage plans. It also points to a March 2026 settlement between one payer and the Department of Justice over upcoding allegations, and a lawsuit Massachusetts filed in May 2026 against another large payer.
| Question | BCBSA’s position | AHA’s position |
|---|---|---|
| Why is coding more complex? | AI tools surface billable secondary diagnoses | Older, sicker patients and new coding guidelines |
| Has care changed? | No evidence of a matching change in treatment | Case-mix index up about 5% from 2019 to 2024 |
| What does the AI do? | Scans labs and notes for anything that can be coded | Improves accuracy, with human validation |
| Who is distorting payments? | Hospitals billing for care not delivered | Insurers downcoding claims and upcoding in Medicare Advantage |
| What should happen next? | Clear expectations for hospitals that use AI coding | Stop insurers cutting payment on the pretext of upcoding |
Where the two sides overlap
Strip away the rhetoric and there is common ground. The AHA says human validation “remains essential” to accurate coding, and that hospitals have legal, ethical and contractual obligations to code appropriately. BCBSA says AI can reduce administrative burden “when used appropriately” and wants billed diagnoses to match delivered care. “We should be reimbursing for care delivered,” Chalker said. Neither side disputes that principle. They dispute whether the AI upcoding data shows it being broken.
Who Pays for AI Upcoding: Premiums, Employers and Taxpayers
BCBSA’s framing is about affordability. It says spending on complex coding with no change in care flows through to higher premiums and out-of-pocket costs for families, employers and taxpayers. “As families face higher healthcare costs, this research underscores the urgent need to better understand these AI tools and the role they may play in exacerbating the affordability crisis,” said David Merritt, BCBSA’s senior vice president of external affairs.
How a coding shift reaches a premium
Insurers set premiums from expected claims costs, so a sustained rise in what each hospital stay costs feeds into the next year’s rates. Employers that self-fund their health plans pay claims directly, so they see the effect sooner. BCBSA’s own list of who bears the cost also includes taxpayers.
The issue is not confined to Blue plans. Chalker said other insurers also see coding change and “talk about it sometimes in earnings reports”. UnitedHealth Group, for example, said on its July earnings call that commercial medical cost trend was running modestly above 11%, and named provider coding intensity as one cause, according to an analysis published on Yahoo Finance.
What vendors tell hospitals
The March brief listed public case studies from AI documentation and coding vendors, and they show why hospitals buy. One ambient listening vendor says its product helps hospitals make an additional $13,000 per clinician each year. A coding AI vendor markets a 5:1 return on investment “on day one”. Another customer announcement cites a 4% increase in coding complexity through more precise documentation, and a fourth reports a 5.1% revenue uplift.
From the hospital side, those are efficiency and accuracy gains. From the payer side, they are the exact costs the AI upcoding reports are trying to measure.
Adoption is still climbing
BCBSA’s March summary cited a 2023 HFMA and AKASA survey in which 46% of hospitals and health systems used AI in billing, coding and claims. It also cited federal data showing 7 in 10 US hospitals used predictive AI in 2024, with AI use for billing up 25 percentage points in a year. By September, BCBSA was citing more than 60% of hospital systems. The white paper’s conclusion is blunt: “this growth hasn’t stopped, nor is it limited to inpatient care.” If the AI upcoding reports are right, the cost will keep climbing with adoption.
What Comes Next for AI Upcoding Oversight
The September paper builds on the March study, and BCBSA says more AI upcoding analyses are coming. BCBSA says its data, drawn from plans covering one in three Americans, gives it “the unprecedented ability to identify these patterns, measure impact and proactively address cost drivers across the healthcare system.”
More analyses, and a push on payment rules
Chalker said BCBSA plans further releases on outpatient care and other DRG families. The association says it is supporting Blue plans in using data to spot upcoding trends and in “establishing clear expectations for hospitals using AI coding tools to ensure billing matches delivered care”. The March brief went further, warning that payment models “will increasingly reward more intensive coding unless Plan safeguards or adjustments to the DRG payment system keep pace.”
In practice, that points towards more clinical validation reviews of lab-derived bump codes and contract terms that tie coded severity to treatment. It is also what the AHA warns against: automated payment cuts justified by claims of AI upcoding.
What hospitals and health IT teams should do now
For any organisation running AI documentation or coding tools, the practical defence against an AI upcoding audit is to run the insurers’ test before the insurers do. The same principles apply to any clinical AI rollout, as covered in our guide to machine learning in healthcare.
- Track coded severity against treatment. Compare how often a lab-derived diagnosis is coded with how often it is treated, such as anemia against transfusions.
- Set clinical thresholds for lab-derived codes. Do not let a tool suggest a diagnosis for every out-of-range value that no clinician acted on.
- Keep a human decision on every suggestion. Log which AI-proposed codes coders accepted and rejected, and why.
- Benchmark against peers. A hospital whose complexity rises much faster than similar hospitals will be the first one audited for AI upcoding.
- Keep the clinical evidence with the claim. If a diagnosis changed the payment, the record should show why it mattered to care.
None of this assumes the insurers are right about AI upcoding. It simply means a hospital can show its coding reflects care, which is the one standard both sides already accept.
AI Upcoding FAQ
What is AI upcoding?
AI upcoding is the allegation that AI-powered documentation and coding tools lead hospitals to bill for more severe diagnoses than the care delivered supports, pushing claims into higher-paying categories. Upcoding itself long predates AI. The question now is whether software makes it faster and more systematic.
How much does the BCBSA report say AI upcoding cost?
About $942 million to Blue Cross and Blue Shield companies over 2024 and 2025, measured against 2023 coding rates. About $653 million of that came from secondary diagnoses that bumped claims into higher-severity DRGs, at an average of $11,800 per excess complex case.
Does the report prove AI caused the increase?
No. It shows coding complexity rising without a matching rise in treatment, concentrated in a subset of hospitals, during a period of fast AI adoption. The link to AI upcoding is an inference from timing and from the type of codes involved. BCBSA itself notes the analysis relies on claims rather than clinical records.
What do hospitals say?
The American Hospital Association calls the AI upcoding claims unsubstantiated. It says patients are older and sicker, that coding guidelines have changed, and that AI tools improve accuracy under human oversight. It also accuses insurers of automated downcoding and of upcoding in their own Medicare Advantage business.
Why do some reports say $2.3 billion?
That figure comes from BCBSA’s March 2026 study, which estimated about $663 million in inpatient and at least $1.67 billion in outpatient spending nationwide. The $942 million figure is from the September white paper and covers inpatient costs to Blue plans only.
References
BCBSA Analysis: How AI Coding Tools Affect Healthcare Costs (Blue Cross Blue Shield Association)
Hospital Coding Intensity Analysis: Major Bowel Procedures (BCBSA white paper)
New BCBSA Research Shows AI Billing Raises Health Care Costs (Blue Cross Blue Shield Association)
Studies Show How AI Is Shaping Hospital Billing Trends (Blue Cross Blue Shield Association)
Rising Coding Intensity and Its Impact on Health Care Affordability (Blue Health Intelligence)
Fact Sheet: Artificial Intelligence and Coding Intensity (American Hospital Association)
Hospitals’ use of AI coding tools cost BCBSA plans $942M more for similar care (Fierce Healthcare)
AI tools for health generated nearly $1 billion in extra costs: report (The News)
UnitedHealth: Inside the $942 Million Industry Problem Squeezing Its Margins (Yahoo Finance)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
