Healthcare costs for American employers are projected to rise at their fastest pace in nearly two decades, and the country’s health insurers now say artificial intelligence is one of the reasons. On 26 September 2026, TechCrunch summed the argument up in a headline: insurers claim AI is already increasing healthcare costs. The trigger was an analysis from the Blue Cross Blue Shield Association (BCBSA) estimating that more complex hospital billing, which it links to AI coding tools, added $942 million to what Blue plans paid over two years.
The claim is not coming from one trade group alone. In June, most of the health plans surveyed by PwC named providers’ AI documentation and coding tools as a top-three driver of next year’s rise in healthcare costs. In April, the Peterson Health Technology Institute reported that AI scribes were lifting the billing level of routine visits. On 23 September, Dr Mehmet Oz, who runs the Centers for Medicare and Medicaid Services (CMS), said AI would be “inflationary” in the short term. Hospitals reply that they are finally being paid for care they already deliver, and that insurers run AI of their own.
This article takes each insurer claim in turn: what was measured, who measured it and what it can and cannot show. It covers how AI changes a bill on both sides of a claim, the hospital rebuttal, the insurers’ own coding record, the likely scale of the effect on healthcare costs and the fixes now on the table. Our earlier breakdown of the BCBSA white paper and AI upcoding covers the inpatient coding mechanics in more depth.
Table of contents
- What Insurers Claim About AI and Healthcare Costs
- Four Sources Behind the Healthcare Costs Warning
- How AI Tools Push Healthcare Costs Up, Claim by Claim
- Insurers Use AI Too: The Other Side of Healthcare Costs
- Bots Fighting Bots: Why Healthcare Costs Can Rise on Both Sides
- The Hospital Rebuttal on Healthcare Costs
- Insurers’ Own Coding Record and Healthcare Costs
- How Big Is the AI Effect on Healthcare Costs?
- What Oz, Regulators and Plans Want to Do About Healthcare Costs
- Timeline: AI and Healthcare Costs in 2026
- What Rising Healthcare Costs Mean for Patients and Employers
- Lessons for Teams Building AI That Touches Healthcare Costs
- Healthcare Costs and AI: Frequently Asked Questions
- References and Further Reading
What Insurers Claim About AI and Healthcare Costs
The insurer case rests on one pattern: bills are describing sicker patients, while the treatment those patients receive looks much the same. Insurers say AI is the new ingredient that explains the gap, and that the gap is now large enough to show up in healthcare costs.
The TechCrunch report and the $942 million estimate
TechCrunch’s Anthony Ha built the story on two pieces of reporting. The first is BCBSA’s analysis, released on 24 September, which found “a sharp increase in patients being documented as having complex conditions” with “no evidence of corresponding change in care delivered”. The second is a New York Times report from the same day, which treated the Blue Cross numbers as the latest sign that AI is contributing to rising healthcare costs and described AI now being used on both sides of the old fight between hospitals and insurers.
The core figures are simple. According to BCBSA, the share of inpatient cases billed as medically complex rose from about 37% at the start of 2023 to about 40% by the end of 2025. The association prices that drift at $942 million in extra healthcare costs for Blue plans between 2023 and 2025. About 70% of it, roughly $650 million, came from secondary diagnoses that moved a stay into a higher-paying category. More than 60% of hospital systems now use AI coding tools, the association says.
“It’s not a war. It’s a completely one-sided blood bath”
Luke Chalker, a BCBSA senior vice president and one of the analysis’s authors, framed the finding in clinical terms. “If patients are truly sicker, we’d expect to see more treatment,” he said in the association’s release. “The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.”
Quoted in the Times report, he rejected the idea that hospitals and insurers were fighting an even battle over AI billing. “It’s not a war. It’s a completely one-sided blood bath,” he said, with insurers on the losing side.
That line is advocacy, and it should be read as advocacy. BCBSA speaks for the companies that pay the claims, and its analysis relies on claims data rather than patient charts. But the same direction of travel shows up in work that insurers did not write, which is why the argument about AI and healthcare costs moved from trade-press rebuttals to the head of CMS in a single week.
How this piece differs from the white paper coverage
Our AI upcoding article walks through the BCBSA white paper itself: the diagnosis-related group tiers, the bowel surgery case study and the anemia test. This piece steps back. It asks whether the wider claim, that AI is already increasing healthcare costs across the system, holds up once every source is lined up and the insurers’ own tools are counted as well.
Four Sources Behind the Healthcare Costs Warning
Four separate bodies have made some version of the insurer argument in 2026. They use different data, measure different parts of the system and have different interests in the answer. The table puts them side by side before each is examined.
| Source | Date | What it says | Evidence base | Headline figure |
|---|---|---|---|---|
| Blue Cross Blue Shield Association | 24 Sep 2026 | Hospital AI coding adds cost without adding care | Blue plan inpatient claims, 2023 to 2025 | $942 million over two years |
| PwC Health Research Institute | 11 Jun 2026 | Provider AI is a top-three cost inflator for 2027 | Survey of actuaries at 27 health plans | Nearly 70% of plans agree |
| Peterson Health Technology Institute | 13 Apr 2026 | AI scribes and coding tools raise billing intensity | Workshop of systems, plans, vendors and agencies | $1,004 per provider per month |
| CMS Administrator Dr Mehmet Oz | 23 Sep 2026 | AI will be inflationary before it saves money | Public remarks at an industry summit | No figure given |
Blue Cross Blue Shield Association, September 2026
BCBSA’s analysis is built from de-identified claims from Blue companies that together cover one in three Americans. Its three-page white paper counts 55,158 extra complex cases compared with 2023 rates, worth $653 million, and uses major bowel surgery as its case study. Both studies point at hospital AI tools as the likely cause. It follows a March study of maternity care in which one diagnosis, acute posthemorrhagic anemia, added an estimated $22 million to maternity admission costs in a single year.
David Merritt, the association’s senior vice president of external affairs, tied the work directly to affordability. “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,” he said.
PwC’s survey of 27 health plans, June 2026
The broadest insurer-side evidence comes from PwC’s Behind the Numbers 2027 report. PwC interviewed actuaries at 27 plans covering more than 103 million employer-sponsored members and 8 million individual marketplace members. They projected a 9% group medical cost trend for 2027 and 8.5% for the individual market, the highest in nearly two decades. They also restated 2026 upwards, from 8.5% to 9% for group plans.
Nearly 70% of plans ranked providers’ AI documentation and coding tools as a top-three inflator of healthcare costs, and about 20% called it the single largest one. Glenn Hunzinger, PwC’s US health industries leader, put that in proportion. “It does have an impact on that 9%, albeit it’s not the biggest piece,” he told Healthcare Dive.
The chart matters because it shows how the insurers themselves rank the problem. AI appears near the top of their worry list, but pharmacy spending sits above it. Only a minority of plans think AI coding is the biggest single force behind healthcare costs.
Peterson Health Technology Institute, April 2026
The Peterson Health Technology Institute (PHTI) is a nonprofit founded in 2023 to evaluate health technology. In January 2026 it convened senior leaders from health systems, health plans, technology developers, investment firms and federal agencies. Its April report on administrative AI is blunt: “Provider deployment of AI is increasing billing intensity and inflating medical spending.”
The report cites one multihospital system where an AI scribe was followed by a 5% rise in the highest-level (level 5) visits and a 7% rise in level 4 visits for established patients. That was worth an average of $1,004 per provider per month. PHTI’s executive director, Caroline Pearson, told STAT what participants said behind closed doors: “it’s quite clear scribes are increasing coding intensity. One hundred percent.”
CMS Administrator Mehmet Oz, September 2026
The newest voice is the most senior. Speaking at Oracle’s annual health and life sciences summit in Orlando on 23 September, Oz gave what Healthcare Dive called a candid assessment: “Short term, AI is going to be inflationary because it’s going to turbocharge the ability of the current billing systems to work more effectively.”
He did not suggest slowing down. “I guarantee you we will lose lives if we don’t use AI in the day-to-day trench warfare of fighting disease in America,” he said. His long-term hope for bringing healthcare costs down rests on accountable care organisations, which are covered later in this article.
How AI Tools Push Healthcare Costs Up, Claim by Claim
None of the insurer claims says AI invents patients or procedures. The mechanism is documentation. US hospitals and doctors are paid according to codes, and codes depend on what is written in the record. A tool that writes more of the record, or reads it more thoroughly, can move a bill into a higher band without any change at the bedside.
Inpatient stays: one extra diagnosis, one higher tier
Hospital stays are paid by diagnosis-related group. Many groups have up to three price tiers: no complication, a complication or comorbidity (CC), or a major one (MCC). A single secondary diagnosis on the record can lift the whole stay a tier. BCBSA’s example is anemia after major bowel surgery, which can be derived from a single lab value and is therefore easy for software to spot.
In 2025 bowel procedure cases at the quarter of hospitals where complex coding grew most, anemia appeared in 13.7% of cases against 9.9% elsewhere. Transfusion rates among those anemia patients were lower, 16.9% against 19.3%. Dividing the white paper’s $653 million by its 55,158 extra complex cases gives about $11,839 per case, roughly the average price of one step up the ladder.
Outpatient visits: the evaluation and management ladder
Office and clinic visits are billed with evaluation and management (E/M) codes. New patients run from 99202 to 99205 and established patients from 99211 to 99215. Higher levels pay more. Since a 2021 revision, the level depends mainly on the complexity of medical decision-making or on total time, rather than on the length of the history and examination write-up.
Trilliant Health studied six large health systems that had announced ambient AI scribing, serving 13 states between them, using all-payer claims from 2018 to 2024. Every one shifted towards the two highest levels. The established-patient figures are below.
| Health system | 99214-99215 share, 2018 | 99214-99215 share, 2024 | Change |
|---|---|---|---|
| System A | 59.8% | 67.2% | +7.4 points |
| System B | 40.9% | 52.7% | +11.8 points |
| System C | 65.3% | 72.9% | +7.6 points |
| System D | 50.4% | 60.2% | +9.8 points |
| System E | 56.6% | 64.6% | +8.0 points |
| System F | 47.8% | 57.7% | +9.9 points |
New-patient visits moved further still. The chart shows the rise in the share billed at 99204-99205 at each system, computed from Trilliant’s 2018 and 2024 figures.
Trilliant is careful about causes. Its data start in 2018, before scribe adoption accelerated, and the shift coincides with the 2021 guideline change. Its conclusion is that the pattern “warrants examination but likely reflects enhanced rules-based documentation”, not fraud.
Ambient scribes capture more of the visit
Ambient scribes listen to a consultation and draft the clinical note. Trilliant describes them as using natural language processing and a large language model to structure the conversation into the record, and some platforms also suggest billing codes. The benefit to clinicians is real: in one study of 263 providers across six systems, burnout fell from 51.9% to 38.8% after 30 days with a scribe.
The billing effect follows from completeness. A scribe records every symptom, every chronic condition mentioned in passing and the full length of the visit. PwC points to a University of California, San Francisco study in which scribe use was associated with higher relative value units per encounter and no measurable increase in claim denials. In other words, the extra billing was paid.
Vendors paid on what they find
Some coding tools are sold on the revenue they surface. SmarterDx, which reviews charts before a bill goes out, advertises a 5:1 return and an average of $2.5 million in realised annual net new revenue per 10,000 discharges. It says it finds new CC or MCC opportunities in 36% of encounters. Its case study with McLaren Health Care, a 2,246-bed Michigan system, reports more than $11 million a year.
According to the Times, McLaren’s chief financial officer, Dave Mazurkiewicz, said the tool has added about $1 million a month, and SmarterDx takes a cut of what it finds. In September 2025 Smarter Technologies, the company behind SmarterDx, bought Pieces Technologies and launched SmarterNotes, which drafts the clinical note with reimbursement logic built in. For insurers, that merger of note-writing and billing is exactly the worry. It is also why the insurer case on healthcare costs points at vendor incentives as much as at hospitals. Our overview of AI medical coding software explains how these products fit into the revenue cycle.
Insurers Use AI Too: The Other Side of Healthcare Costs
The insurer framing leaves out half of the picture. Health plans have used automated claim review for years, and they are adding AI to it quickly. Every one of those systems is designed to pay less on some claims, and every one creates work on the provider side that feeds back into healthcare costs.
Payment integrity moves before the payment
PwC’s own advice to plans is to move review earlier. “Payment integrity becomes less about post-payment recovery and more about confirming the validity of high-risk claims before payment is released,” its researchers wrote. They list high-dollar inpatient claims, DRG shifts, modifiers and implant charges as priorities. “The goal is not more denials; it is more accurate payment before dollars leave the plan.”
That is a reasonable goal. It also describes a machine built to examine exactly the claims that hospital AI is built to produce.
Downcoding algorithms
PHTI found that plans are already responding to AI-driven billing with “across-the-board downcoding and other reimbursement reductions”. Reported tactics include algorithms that automatically reduce outlier high-complexity E/M codes, lower payment for certain modifiers and compare each doctor’s billing with their peers. Several plans paused, delayed or withdrew those policies after provider opposition and state regulatory scrutiny.
The American Hospital Association says these “downcoding programs” often use automated edits to cut payment “without reviewing medical documentation”, forcing providers into appeals that the plan then frequently overturns. PHTI adds a sharper warning: blanket cuts may hurt providers who never adopted AI at all, who tend to be smaller, rural, critical access and independent.
Denials at machine speed
The best-documented insurer automation predates generative AI. In 2023, ProPublica reported that Cigna’s PxDx system let its medical directors deny more than 300,000 payment requests over two months in 2022, spending an average of 1.2 seconds on each. That works out at about 100 hours of review in total. Cigna called the description of the process “incorrect”.
The same year, a class action alleged that UnitedHealth used an algorithm called nH Predict to cut off rehabilitation care for Medicare Advantage patients. The complaint, reported by STAT, claimed a 90% error rate based on reversed denials. UnitedHealth said the tool is not used to make coverage decisions and that the suit has no merit.
Medicare’s own AI reviewer: WISeR
The federal government is now building the same machinery. The WISeR model uses AI and machine learning, with human clinical review, to check medical necessity for selected services in traditional Medicare. It runs from 1 January 2026 to 31 December 2031 in six states: New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington. Technology companies are the only participants, and the targets include skin substitutes, nerve stimulator implants and knee arthroscopy for osteoarthritis.
| Stage of a claim | Hospital or doctor side | Insurer side | Effect on healthcare costs |
|---|---|---|---|
| During the visit | Ambient scribe drafts a fuller note | None directly | More billable detail recorded |
| Before care is given | AI drafts prior authorisation requests and justifications | AI triages requests and supports decisions; WISeR in Medicare | Faster steps, but more requests and appeals |
| Before the bill goes out | Pre-bill review finds missed diagnoses | None directly | Higher tiers and levels billed |
| Before payment | None directly | Pre-payment review and downcoding edits | Lower payments, more disputes |
| After a denial | AI drafts appeals | Automated review of appeals | More rounds, each one cheaper |
Bots Fighting Bots: Why Healthcare Costs Can Rise on Both Sides
Put the two halves together and a different picture emerges. The question is no longer whether hospitals or insurers are using AI. Both are. The question is whether two sets of automated systems arguing over the same claim make care cheaper or simply make the argument faster.
Rao’s warning
Dr Shiv Rao, a cardiologist and the founder of the ambient scribe company Abridge, gave the Times the phrase that framed the week. AI on both sides could lead to “a horrible dystopic future nobody wants to live in,” he said, with “bots fighting bots, agents fighting agents.” He also argued that the same technology could reduce tension and cut healthcare costs. Which outcome arrives depends on what the systems are built and paid to do.
Cheaper rounds mean more rounds
PHTI’s report explains why automation on both sides may not lower healthcare costs. In prior authorisation, AI lets providers submit more complete requests with less effort, and lets plans process more of them at a lower cost per decision. But PHTI found “no evidence yet that this translates to lower average cost per claim factoring in the cost of the AI solution.”
Its participants worried that “optimizing each side of the transaction risks making the overall process more activity-intensive, rather than more efficient.” When a round of argument costs almost nothing, both sides can afford more rounds.
A data centre full of arguments
The Times also collected the sharpest image of that loop. According to a summary of its report by Business Model Analyst, Emory Healthcare’s chief executive pictured his system’s bots and Cigna’s bots trading arguments until they filled a data centre. Cigna and Emory also shared a stage in New York to talk about a truce.
The image is a warning about incentives, not a forecast. Each side’s system can report a strong local return while the combined healthcare costs of the process rise. For patients, whose premiums fund both sides of the dispute, that combined figure is the one that matters.
Administrative healthcare costs are both the prize and the risk
The American Hospital Association puts US administrative spending at more than $1 trillion a year. CMS says waste accounts for up to 25% of US healthcare spending. Cutting either would lower healthcare costs by far more than the sums now in dispute, and that is the promise every AI vendor makes. The insurer claim, read generously, is that the first wave of AI has added a layer to that administrative stack rather than removing one.
The Hospital Rebuttal on Healthcare Costs
Hospitals reject the insurer story. Their case is that coding is catching up with reality, that patients are sicker than they were and that insurers are using the AI charge as cover for cutting payments. Each part of that case has evidence behind it.
Sicker patients and new coding rules
The American Hospital Association’s July fact sheet calls the insurer claims “unsubstantiated”. It says 19% of hospital expense growth from 2019 to 2024 reflects caring for sicker, more complex patients. An AHA and Vizient analysis found hospital case-mix index, a standard measure of how ill inpatients are, rose by about 5% over the same period.
The AHA also points to the continued shift of simpler care to outpatient settings, which leaves hospitals with a sicker inpatient population. It notes that changes in E/M guidelines and diagnostic coding systems have themselves pushed reported acuity upwards.
Years of undercoding
Trilliant Health’s study makes the undercoding argument explicitly. Because coding is rules-based, a systematic rise could mean that doctors “historically under-coded”, through incomplete notes, weak coding practice or fear of the False Claims Act. It calls the shift “unlikely attributable to provider fraud”, noting that fraud on that scale would need coordination across unrelated systems.
Even PwC’s Hunzinger accepts part of this. Many health systems “were probably missing correct codes” because of the volume and complexity of their own systems, he told Healthcare Dive. Thin margins and cuts to Medicaid give them every reason to stop missing them.
“They want it both ways”
The AHA’s sharpest point concerns consistency. Many of the insurers now running downcoding programmes have told investors that their own members are sicker and need more services. “In short, these plans want it both ways,” the fact sheet says: a sicker population for their risk scores, and a healthier one when it is time to pay hospital claims.
| Question | Insurer claim | Hospital reply | What would settle it |
|---|---|---|---|
| Are patients sicker? | Treatment has not changed, so no | Case-mix index is up about 5% | Chart review of a matched sample |
| Is the coding accurate? | Codes are chosen because they pay | Codes follow official guidelines | Independent coding audits |
| Is AI the cause? | Growth tracks AI adoption | Growth began before AI and follows rule changes | Comparison of adopters with non-adopters over time |
| Was there undercoding before? | Not addressed | Yes, notes were incomplete | Scribe transcripts checked against codes |
| Who is gaming the system? | Hospitals and their vendors | Insurers, through risk scores and downcoding | The same audit standard applied to both |
Insurers' Own Coding Record and Healthcare Costs
The hospital counterattack lands because insurers have a coding problem of their own. In Medicare Advantage, the government pays private plans more for members who are recorded as sicker. That creates the same incentive to document every diagnosis, and federal auditors keep finding that plans go too far. Any honest account of AI and healthcare costs has to include it.
OIG audits of Humana and UnitedHealthcare
On 15 September, the HHS Office of Inspector General published audits of two plans run by the largest Medicare Advantage insurers. Healthcare Dive reported that HumanaChoice received an estimated $130.9 million and UnitedHealthcare of Wisconsin $46.9 million in improper payments for 2020 and 2021. Medical records failed to support 178 of 220 sampled enrollee-years at Humana and 183 of 250 at UnitedHealthcare.
Both insurers disputed the methodology and said they would not return the money. Earlier in the year, the OIG found problems at Blue Cross and Blue Shield of Alabama ($7 million), Gateway Health Plan ($4.3 million) and Priority Health ($4.4 million) for 2018 and 2019. One of those is a Blue plan.
The Medicare Advantage bill
The sums are larger than anything in the hospital dispute over healthcare costs. Healthcare Dive reports that the US will spend an estimated $76 billion more on Medicare Advantage this year than traditional Medicare would cost, partly because of upcoding. The AHA cites MedPAC’s 2025 finding that upcoding contributed $40 billion in excess payments to plans.
The AHA also points to a March 2026 Justice Department settlement with Aetna for $117.7 million and a May 2026 lawsuit by the Massachusetts attorney general accusing UnitedHealthcare of defrauding MassHealth of about $100 million. Trilliant cites a Senate report documenting software used to “manufacture” higher-acuity diagnoses for Medicare Advantage payments.
Oz’s garden
Oz has noticed. At a Better Medicare Alliance event in Washington on 22 September, a day before his AI remarks, he compared Medicare Advantage to a garden. “In recent years, our garden has laid vulnerable to weeds and overgrowth,” he said. “And for too long, patients have paid the price both financially and physically.” His Medicare director, John Brooks, spoke of “a crisis of confidence” in the programme.
The chart does not add these figures together, because they cover different payers, years and methods. It shows that both sides of the dispute have been accused of coding more than the care justifies, with sums of broadly similar scale. The Medicare Advantage estimates dwarf all of them.
How Big Is the AI Effect on Healthcare Costs?
A $942 million headline sounds enormous. Set against total US healthcare costs it is small, and that context matters for judging the insurer claim fairly. The figures below use only numbers stated in this article and in the CMS national health expenditure data.
Putting $942 million in scale against national healthcare costs
CMS reports that national health spending grew 7.2% to $5.3 trillion in 2024, or $15,474 per person. Private health insurance spending grew 8.8% to $1,644.6 billion, and hospital spending grew 8.9% to $1,634.7 billion.
| Calculation | Working | Result |
|---|---|---|
| BCBSA estimate per year | $942m / 2 years | $471m a year |
| Share of 2024 private insurance spending | $471m / $1,644.6bn | About 0.03% |
| Share of 2024 hospital spending | $471m / $1,634.7bn | About 0.03% |
| Extra cost per complex case | $653m / 55,158 cases | About $11,839 |
| PHTI scribe effect per doctor per year | $1,004 x 12 months | $12,048 |
| OIG estimates, two MA plans combined | $130.9m + $46.9m | $177.8m over two years |
| Cigna PxDx review time | 300,000 x 1.2 seconds | About 100 hours |
On its own, then, the BCBSA figure is a rounding error in national healthcare costs. But it is deliberately narrow. It covers only Blue plans, only inpatient stays, only the categories studied and only cases where BCBSA found no matching change in care. The outpatient shift that Trilliant and PHTI describe is a separate, and potentially larger, effect.
A driver of healthcare costs, not the largest one
PwC’s survey helps with proportion. Health plans see AI coding as a real inflator but not the biggest. Pharmacy spending, especially on specialty drugs and GLP-1 medicines, ranks higher. Provider consolidation and contracting pressure sit alongside it. Behavioural health use rose 10% from 2023 to 2024 and 62% since 2018, according to Trilliant data cited by PwC.
Disputes under the No Surprises Act add more. PwC says providers won 88% of the 2.6 million independent dispute resolution cases filed in 2025. On insurer-side evidence alone, AI is one pressure on healthcare costs among several, which is what Hunzinger said.
What the evidence cannot show yet
Three gaps remain. First, most of the insurer evidence uses claims, not charts, so it cannot show directly whether patients were sicker. BCBSA acknowledges this. Second, the link to AI is inferred from timing and adoption rather than from a comparison of hospitals that did and did not use the tools. Trilliant’s six systems were chosen because they announced scribes, so there is no control group. Third, nobody yet knows the net effect of insurer downcoding, which PHTI says is “not yet known”. Until those gaps close, every estimate of AI’s effect on healthcare costs is a bound, not a measurement.
What Oz, Regulators and Plans Want to Do About Healthcare Costs
If AI is pushing healthcare costs up through documentation, the fixes fall into three groups. Change how care is paid for, adjust prices for coding drift, or make the tools themselves more transparent. Each has backers, and each has costs.
Oz’s bet on accountable care to lower healthcare costs
Oz’s long-term answer is accountable care organisations (ACOs). These are groups of providers that take responsibility for the total cost and quality of care for a defined population, such as Medicare members. The idea is that when providers are paid for outcomes rather than for each billed service, AI becomes a tool for efficient care rather than an engine for generating bills.
CMS has launched several programmes to increase ACO participation and says they save Medicare billions. Critics warn that ACOs put financial risk on providers who fail to lower spending and add another layer of administration.
A coding adjustment, Medicare-style
Medicare has been here before. When it introduced severity-based MS-DRGs in fiscal 2008, CMS expected hospitals to improve their coding and built documentation and coding adjustments into its rates. Congress later ordered CMS, through the American Taxpayer Relief Act of 2012, to recover $11 billion tied to that coding shift, which it did through rate cuts from fiscal 2014 to 2017.
Commercial contracts have no such mechanism. PwC tells plans they will need “targeted protections against reimbursement drift once contracts are in place”. A coding-intensity clause priced at contract renewal would be the private equivalent of Medicare’s adjustment. Like Medicare’s version, it would cut accurate and aggressive coders alike.
Disclosure, audits and price resets
PHTI calls reimbursement policy “the strongest lever” and says current plan responses are “likely not sufficient to address AI-driven medical inflation”. Its participants backed near-term guardrails to detect what is driving inflation, and “generally supported adjusting the price of medical services” in response to AI.
Trilliant suggests a simpler route: because scribes record the whole encounter, payers can compare the documentation with the code and check the match. Disclosure of which AI coding tools a provider uses would make that easier.
Real-time prior authorisation
On the insurer side, the most promising idea is to settle prior authorisation during the visit instead of arguing about it for days. PHTI cites self-reported results from Optum Rx’s PreCheck prior authorisation pilot with Cleveland Clinic: 88% fewer appeals and 68% fewer denials caused by missing information.
A CMS rule, CMS-0057-F, requires Medicare Advantage, Medicaid, CHIP and federal marketplace plans to offer a prior authorisation API by 1 January 2027. PHTI warns that standard data exchange will not remove the differences in each plan’s medical necessity criteria.
| Proposal | Backed by | What it changes | Main risk |
|---|---|---|---|
| Accountable care organisations | CMS, Dr Oz | Pays for outcomes, weakening the link between codes and income | Financial risk and admin load for providers |
| Coding-intensity price adjustment | Medicare precedent, PwC advice | Offsets drift across all claims at once | Penalises accurate coders too |
| Across-the-board downcoding | Some health plans | Cuts high-level codes automatically | Hurts smaller, non-AI providers; state pushback |
| Documentation-to-code audits | Trilliant Health | Checks each code against the scribe record | Needs data sharing and privacy safeguards |
| Disclosure of AI coding tools | PHTI workshop participants | Lets plans and regulators see where tools are used | Disclosure alone changes no prices |
| Real-time prior authorisation | Plans, vendors, CMS-0057-F | Settles approval during the visit | Pilots are narrow; criteria still differ by plan |
Timeline: AI and Healthcare Costs in 2026
The argument did not start with the TechCrunch headline. It has built through the year, with each new study cited by the next. The dates below come from the sources linked in this article.
| Date | Event |
|---|---|
| January | PHTI convenes health systems, plans, vendors, investors and agencies on administrative AI |
| 1 January | CMS WISeR model starts in six states |
| 12 March | Trilliant Health publishes its study of AI scribing and outpatient coding |
| March | BCBSA maternity study links AI billing to higher costs |
| 8 April | STAT reports that insurers and providers privately agree scribes raise costs |
| 13 April | PHTI publishes “Administrative AI: Current Use and Potential Impact” |
| 11 June | PwC projects a 9% cost trend for 2027, with AI a top-three inflator for most plans |
| 31 July | AHA fact sheet calls the AI coding claims unsubstantiated |
| 15 September | OIG audits of HumanaChoice and UnitedHealthcare of Wisconsin published |
| 22 September | Oz describes Medicare Advantage as a garden overrun by weeds |
| 23 September | Oz says AI will be inflationary in the short term |
| 24 September | BCBSA publishes the $942 million analysis; the New York Times reports on AI on both sides |
| 26 September | TechCrunch: “Insurers claim AI is already increasing healthcare costs” |
Read in order, the timeline shows the insurer claim hardening from private agreement in April to a public headline in September. It also shows the counterclaims arriving in the same weeks, which is why the debate over healthcare costs now has two well-documented sides.
What Rising Healthcare Costs Mean for Patients and Employers
Most patients will never see the argument between two billing systems. They will see its result in their own healthcare costs: premiums, deductibles and the occasional surprise denial. That is why the insurer claim about healthcare costs matters beyond the industry.
Premiums, self-funded plans and healthcare costs
BCBSA says the coding shift is “driving higher premiums and out-of-pocket costs for families, employers and taxpayers”. The mechanism is indirect but real: plans set next year’s premiums from this year’s growth in healthcare costs, and PwC’s 9% trend is exactly the figure actuaries use. Employers that self-fund their plans pay claims from their own budgets, so a higher average cost per stay reaches them directly.
The other side also passes through. Downcoding disputes and appeals cost hospitals money, and those costs are recovered in the prices hospitals negotiate at the next contract.
Smaller providers caught in the middle
PHTI’s warning about non-adopters deserves attention. If plans respond to AI-driven billing with blanket cuts, the providers hit hardest may be the ones that never used the tools, often small, rural or independent practices. They would absorb lower payments without the higher coding that prompted them, which could push more of them towards consolidation and, in turn, higher prices.
Patients and the claim they never see
For patients, more automated denials mean more delays and more appeals to navigate. The nH Predict complaint alleged that the insurer knew only 0.2% of patients would appeal. Automation on the insurer side works best when few people push back. Our look at how AI is reshaping the work of human doctors covers the clinical side of that shift, and our piece on why insurance claims adjusters dislike AI shows the same tension inside insurers.
Lessons for Teams Building AI That Touches Healthcare Costs
For anyone building or buying AI for clinical documentation, coding or claims, this dispute is a preview of the scrutiny to come. The lessons apply to any payment system that pays more for documented complexity. They also fit into a wider AI strategy and IT governance programme rather than a single tool purchase.
Keep an audit trail for every suggestion
Every code or diagnosis a model suggests should be logged with the evidence it relied on and the human who accepted it. Trilliant’s point cuts both ways: scribes create a record that can defend a claim, but only if the link from transcript to code is preserved. Electronic health records already hold most of what is needed.
Separate documentation quality from revenue targets
Tools that are paid on the revenue they find will be judged as revenue tools. Measure documentation accuracy against independent coding audits, and report revenue effects separately. That makes it much easier to answer an insurer’s challenge, or a regulator’s, about why coding intensity rose.
Test against treatment signals
BCBSA’s method is simple enough to copy internally. If a diagnosis becomes more common, check whether the treatment that should follow it also rises. A growing gap between diagnoses and treatment is an early warning. Good data management and analytics make that test routine rather than a one-off project.
Assume the other side runs a model too
Any claim your system produces will be read by another system. Build for that: clean evidence, consistent coding and appeals that cite the record precisely. The same logic applies to insurers designing review models, which need defensible rules and real human review to meet compliance expectations and avoid the lawsuits described above.
Measure net healthcare costs, not local savings
The PHTI finding is the key one for buyers. A tool that saves one side time can still raise total healthcare costs if it triggers more disputes. Ask vendors for evidence of lower total cost per claim, including the cost of the tool, not just faster steps. Our guide to machine learning in healthcare covers how to evaluate those claims.
Healthcare Costs and AI: Frequently Asked Questions
Is AI really increasing healthcare costs?
The evidence points that way for billing, though the size of the effect is disputed. BCBSA, PwC’s surveyed plans, PHTI and CMS Administrator Oz all say AI documentation and coding tools are raising what providers bill. Hospitals say much of the rise reflects sicker patients, new coding rules and past undercoding. None of the studies yet compares adopters and non-adopters directly.
How much has AI added to healthcare costs so far?
There is no system-wide figure. BCBSA estimates $942 million over two years for Blue plans’ inpatient claims alone, about $471 million a year. PHTI cites one health system where an AI scribe added $1,004 per provider per month. PwC’s plans see AI as a top-three inflator of healthcare costs, which they expect to rise 9%, but not the largest one.
Why do hospitals say the insurers are wrong?
The American Hospital Association calls the claims unsubstantiated. It says patients are older and sicker, that case-mix index rose about 5% from 2019 to 2024, and that coding guidelines changed. It also accuses insurers of using automated downcoding to cut payments while arguing that their own members are sicker for Medicare Advantage risk scores.
Do insurers use AI too?
Yes. Plans use automated pre-payment review, downcoding algorithms and AI in prior authorisation. Cigna’s PxDx system and UnitedHealth’s nH Predict have both faced scrutiny, and Medicare’s own WISeR model uses AI for medical necessity review in six states. Abridge founder Shiv Rao has warned this could become “bots fighting bots”.
What did Dr Oz say about AI and healthcare costs?
On 23 September 2026, CMS Administrator Mehmet Oz said: “Short term, AI is going to be inflationary because it’s going to turbocharge the ability of the current billing systems to work more effectively.” He argued that AI is still essential and that accountable care organisations are the route to lowering healthcare costs in the long run.
References and Further Reading
Insurers claim AI is already increasing healthcare costs (TechCrunch)
AI, hospitals and insurers (The New York Times)
BCBSA Analysis: How AI Coding Tools Affect Healthcare Costs (Blue Cross Blue Shield Association)
Hospital Coding Intensity Analysis: Major Bowel Procedures (BCBSA white paper)
Health plans say AI is pushing healthcare costs higher (Healthcare Dive)
Administrative AI: Current Use and Potential Impact (Peterson Health Technology Institute)
Everyone agrees AI scribes are increasing health care costs (STAT)
Outpatient Coding Intensity Following Hospital Adoption of AI-Enabled Scribing (Trilliant Health)
AI will inflate healthcare costs before lowering them, Oz says (Healthcare Dive)
Oz: Medicare Advantage a garden vulnerable to weeds and overgrowth (Healthcare Dive)
Fact Sheet: Artificial Intelligence and Coding Intensity (American Hospital Association)
How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them (ProPublica)
UnitedHealth sued over use of algorithm in Medicare Advantage plans (STAT)
WISeR (Wasteful and Inappropriate Service Reduction) Model (CMS)
National Health Expenditure Fact Sheet (CMS)
McLaren Health Care uncovers over $11M with clinical AI (Smarter Technologies)
Blue Cross Says Hospital AI Cost It $942 Million. Blame the Staircase (Business Model Analyst)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
