This analysis synthesizes 9 sources published the week ending Sep 29, 2026. Editorial analysis by the PhysEmp Editorial Team.
For two years, most coverage of AI in Physician Employment & Clinical Practice has centered on whether software saves clinicians time. Last week the argument moved to money. The Blue Cross Blue Shield Association put a price on hospital AI coding tools, estimating $942 million in added spending from 2023 through 2025, and the CMS administrator said on stage that AI will raise costs before it lowers them. Both sides of the dispute are now running software against each other, and the document they are fighting over is the one physicians sign.
Where the $942 million came from
The BCBSA analysis draws on de-identified claims from plans that cover one in three Americans. Its core finding is a gap between how sick patients were coded and what was done for them. According to Becker’s Payer Issues, the share of inpatient cases coded as complex rose from about 37% in early 2023 to about 40% by late 2025. Roughly $653 million of the total traces to more than 55,000 excess complex inpatient cases, averaging $11,800 each.
Anemia is the association’s showcase example. Hospitals in the top quartile diagnosed it 38% more often than peers yet transfused less, 16.9% of cases against 19.3%. “If patients are truly sicker, we’d expect to see more treatment,” BCBSA’s Luke Chalker said. Secondary diagnoses, often visible in a single lab value, accounted for about 70% of the added cost. More than 60% of hospitals now use AI-enabled coding tools that scan labs, medications, orders and physician notes.
Vendors reject the framing. Codametrix CEO Hamid Tabatabaie told Healthcare Dive that hospitals without such systems were simply missing documented conditions. Travis Bias, deputy chief medical officer at Solventum, whose coding platform is used by more than 80% of U.S. hospitals, pointed at fee-for-service itself: submit more valid codes, get paid more. Both claims can be true at once, which is why this fight will not be settled by a single study.
The study measures hospital claims, but its evidence standard is clinical. Insurers are arguing that a diagnosis without a matching change in treatment is suspect. That test, applied at scale, turns every physician’s problem list into something a payer’s model will try to reconcile against orders.
The chart is where both algorithms meet
Hospital coding software reads what physicians write. Insurer software reads the same record looking for reasons to deny. Hospital executives say they are losing that exchange: HCA’s CFO Mike Marks said hospitals are “behind the payers” on AI claims processing, while Aetna’s chief medical officer, Ben Kornitzer, called the early data “largely inflationary.” Abridge founder Shiv Rao, whose company sells ambient documentation, warned TechCrunch of a future of “bots fighting bots, agents fighting agents.”
For employed physicians, the practical exposure sits in professional billing, and it is already visible in small codes. A Medical Economics piece on the G2211 add-on, worth about $16.05 per qualifying visit, argues that template language generated by ambient AI scribes is drawing payer scrutiny because it lacks patient-specific evidence of a longitudinal relationship. Auditors, the author notes, judge narrative substance, and there are no safe-harbor phrases. Claims pairing G2211 with modifier 25 face automated similarity scoring.
That lands on the physician in two ways. The signature on the note carries the compliance risk regardless of which software drafted it. And the productivity models many employers use count billed work, so a denied add-on or a downcoded visit shows up in wRVUs. A physician group that adopts an ambient scribe to save time may find that the saved minutes go to defending notes that read too much alike.
Fee schedule math that can cut the other way
Speaking at Oracle’s health summit, CMS Administrator Mehmet Oz was blunt: “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 said, as reported by Healthcare Dive. The administration is betting on accountable care organizations, where billing volume matters less, to bend that curve.
A second policy thread matters more for physician pay. Federal officials are weighing whether to reimburse AI diagnostic and clinical tools at 60% to 80% of what clinicians receive for the same work, according to Becker’s Physician Leadership, citing Medscape reporting. Former CMS chief medical officer Lee Fleisher cautioned that adding AI to the budget-neutral physician fee schedule could lower physician pay. Brown’s Ateev Mehrotra argued most AI should be treated as a practice expense, like an EHR.
Budget neutrality turns a technology question into a distribution question. Every dollar the fee schedule pays a software tool is a dollar the conversion factor cannot pay a physician, and the specialties with the most automatable reads would feel it first.
Put the two threads together and physicians face an odd squeeze. Hospital AI appears to be raising total spending through facility coding, while a proposal to pay AI directly could shrink the professional share. Recruiters building offers for radiology, cardiology and other image-heavy fields should expect candidates to ask how their groups are positioned if that proposal advances.
Who pays when the arms race escalates
The cost is already reaching employers and patients. The median proposed premium increase for 2027 is 15%, according to a KFF analysis cited by Yahoo Finance, and PwC analysts attribute a 9% increase in insurers’ medical costs next year to AI-backed billing tools. CBS12 noted that employer benefit costs could rise as much as 11%. Venrock’s Bob Kocher framed the variable that decides the outcome: under fee-for-service, AI raises costs; under value-based contracts, it could lower them.
That split maps directly onto physician employment. Physicians in organizations paid on volume will feel the coding contest through audit pressure, denial rates and documentation mandates tied to specific AI tools. Those in risk-bearing groups have less reason to capture every secondary diagnosis and more reason to use AI to prevent admissions, which changes what a good note is for. Both vendors quoted by Healthcare Dive conceded that real relief depends on the shift away from fee-for-service, a transition that has been underway since 2010.
The number to watch is the complex-case share. If BCBSA’s next update shows it still climbing past 40% while transfusion and treatment rates stay flat, expect insurers to push the anemia-style test down from inpatient claims into outpatient E/M audits, where the documentation belongs to individual physicians rather than hospital coding departments. Employment contracts that assign coding liability, specify which AI tools a physician must use, and state whether denied claims count against productivity will matter far more than they did a year ago.
Sources
AI hospital billing added nearly $1B to healthcare costs over 2 years – Healthcare Finance News
Hospitals say they’re losing the AI billing war — a new BCBS study suggests otherwise – Becker’s Payer Issues
Insurers say AI could add billions in health costs. Billing companies disagree – Healthcare Dive
Insurers claim AI is already increasing healthcare costs – TechCrunch
Why AI Scribe Notes Can Put Your G2211 Reimbursement at Risk – Medical Economics
AI will inflate healthcare costs before lowering them, Oz says – Healthcare Dive
The next big question for physician pay – Becker’s Physician Leadership
The reason your healthcare premiums are skyrocketing: AI – Yahoo Finance
AI Is Fueling a New Battle Over Health Care Costs: Medical Billing, Insurance Premiums, Coding Wars – CBS12