What Oz actually said
Speaking at Oracle's health and life sciences summit in Orlando on September 24, 2026, CMS Administrator Dr. Mehmet Oz gave an unusually candid preview of where he expects AI to take healthcare spending: up, before it goes down. "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," Oz said, as reported by MedTech Dive.
He didn't back off the long view, either: "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." His bet on eventually bending the cost curve rests on a specific lever — Healthcare Dive reported that Oz is counting on accountable care organizations (ACOs) to shift providers off fee-for-service and toward outcomes-based payment, so AI gets used to improve care rather than to maximize what gets billed. CMS backed the point structurally in June 2026 by standing up the Office of Health Technology and Products, a unit dedicated to steering how AI gets implemented across CMS programs.
The timing wasn't a coincidence. Oz's comments landed the same week as a new dataset that gave his warning a dollar figure.
The bill that just landed
The Blue Cross Blue Shield Association released an analysis the same week finding that AI-powered coding tools added nearly $942 million in costs to its member plans between 2023 and 2025, as PYMNTS reported. Of that, $653 million was tied to hospital stays recoded with additional secondary diagnoses, and BCBSA reported that the share of inpatient stays billed as "medically complex" rose from about 37% at the start of 2023 to about 40% by the end of 2025, as more than 60% of hospital systems adopted AI coding tools over that stretch, according to Fierce Healthcare's coverage of the report.
BCBSA's core objection isn't that AI reads charts faster — it's that the additional diagnoses it surfaces don't appear to track with any corresponding change in the treatment patients actually received. Luke Chalker, BCBSA's senior vice president of product and data science, put it plainly: "If patients are truly sicker, we'd expect to see more treatment," as quoted by PYMNTS. That's the shape of AI Oz was describing: it makes an existing billing mechanism more thorough at finding chargeable detail, which raises the bill without changing the underlying care.
Radiology already has a preview of both sides of this
Medical coding isn't the only place "AI" covers very different cost mechanics. Imaging is a useful test case because both patterns Oz described are already visible in how CMS pays for AI there — and they point in opposite directions.
On one side, detection and quantification algorithms are increasingly billed as an additional service layered on top of the scan itself, with their own new code. CMS finalized a Category I CPT code (75577) for AI-driven coronary-plaque quantification and characterization derived from cardiac CT angiography, paying $950.50 in hospital outpatient settings and just over $1,000 in physician offices and imaging centers, effective January 1, 2026, according to AuntMinnie. Separately, CMS approved a New Technology Add-on Payment of up to $137.53 per eligible inpatient case for a CT triage tool that flags suspected acute findings on chest, abdomen, and pelvis scans, available starting with fiscal year 2027 billing on October 1, 2026, per MarketScale. Both are real, useful tools — but structurally, each is exactly the kind of "new billable layer on an existing service" that turbocharges near-term spending the way Oz described.
On the other side sits AI that drafts the radiology report itself. That workflow doesn't touch the CPT code for the scan — a chest CT still bills the same way it did before AI was involved — so there's no new line item for the payer. What changes is internal: how many radiologist-minutes a practice spends producing a report it was already going to bill for.
| Mechanism | Billing layer | Near-term cost direction | Example |
|---|---|---|---|
| Detection / quantification add-on | New CPT or NTAP code, billed on top of the scan | Up — a new charge is added | AI plaque analysis (CPT 75577); CT triage NTAP |
| AI-drafted reporting | None — existing scan CPT code is unchanged | Down — same bill, less radiologist time to produce it | AI generates a structured draft report a radiologist finalizes |
Why the mechanism matters more than the "AI" label
Oz's warning and the BCBSA study both describe AI making an existing revenue-generating process — coding, billing, documentation — more efficient at generating revenue. That's a real and reasonable thing to flag: it's inflationary because it adds detail (and dollars) to a bill without adding care.
Radiologist time is different. It's not a line item a payer sees; it's a fixed, scarce input — one of the reasons the field faces a well-documented radiologist shortage that shows up as longer report turnaround times rather than a bigger bill. AI that reduces the minutes needed to produce a report a practice was already going to generate and bill doesn't turbocharge the billing system Oz is worried about — it turbocharges throughput against a fixed reimbursement, which is the deflationary half of his own long-term prediction, just arriving through a different door than ACOs.
What this means for imaging leaders evaluating AI
Ask which cost the AI is touching
Does the tool add a new billable code, or does it change how much staff time a study takes to produce? Those are different economic bets, and vendors rarely frame the distinction explicitly.
Add-on-code AI still has to prove its clinical case
A new CPT or NTAP payment is a signal a technology cleared CMS's evidentiary bar, not a verdict on whether it belongs in your workflow — the buying decision still comes down to accuracy data and fit for the practice.
Reporting-throughput AI shows up in operating cost, not a new payer line
Its ROI case has to be made in radiologist hours saved and turnaround time, which is a harder sell to a CFO used to add-on payment tables — but it is the mechanism least exposed to the inflation Oz described.
Where xAID fits
xAID's foundation-model-based CT reporting sits in the second column of that table. It doesn't create a new billable service — a client bills the same CPT code for the same scan they always have — it drafts the structured report so a radiologist spends less time producing it. xAID's in-house radiologist reviews every preliminary draft, and the report reaches a client ready-to-sign, with the client's own reading radiologist providing the final signature. That's the version of AI cost-reduction Oz's long-term bet depends on: less time and delay per study the payer was already going to reimburse, not a new line added to the bill.
Frequently asked questions
How does AI reduce costs in healthcare?
AI reduces healthcare costs mainly by cutting the labor time and delay needed to produce a service that's already billed — not by adding a new charge. In radiology, that means AI that drafts a report cuts radiologist minutes per study without creating a new billable line. It's distinct from AI that adds a new, separately billed layer on top of an existing service, such as a detection or quantification add-on with its own CPT code, which raises near-term costs rather than lowering them.
Why did CMS Administrator Mehmet Oz say AI will increase healthcare costs before lowering them?
Speaking at Oracle's health and life sciences summit on September 24, 2026, Oz said AI would be "inflationary" short term because it "turbocharges" existing billing systems to work more efficiently, before an expected long-term drop in costs as care shifts toward outcomes-based models such as accountable care organizations. His remarks came alongside a Blue Cross Blue Shield Association study finding AI coding tools added nearly $1 billion in costs to member plans over two years.
Does AI in radiology always add a new billing cost?
No — it depends on what the AI does. Detection and quantification algorithms increasingly get their own CPT or Medicare add-on code, billed on top of the underlying scan; for example, CMS set a Category I CPT code paying roughly $1,000 for AI coronary-plaque analysis, and a separate add-on payment of up to $137.53 per case for CT triage software. Report-drafting AI works differently: the radiology CPT code for the scan itself doesn't change, so there's no new charge to the payer — only a change in how many radiologist-minutes the practice spends producing that already-billed report.
What did the Blue Cross Blue Shield Association study find about AI and medical coding?
Released September 24, 2026, the BCBSA analysis found AI-powered coding tools contributed close to $1 billion in added costs to its member plans between 2023 and 2025, with $653 million of that tied to hospital stays recoded with more secondary diagnoses. BCBSA said the added complexity wasn't matched by a corresponding rise in treatment delivered.
Source: CMS Administrator Mehmet Oz's remarks at Oracle's health and life sciences summit, as reported by MedTech Dive and Healthcare Dive; CMS Office of Health Technology and Products coverage via Healthcare Dive; Blue Cross Blue Shield Association coding-cost analysis via PYMNTS and Fierce Healthcare; CPT 75577 coronary-plaque payment via AuntMinnie; CT-triage NTAP payment via MarketScale. Figures are rounded as reported.