The Take — Medicare's denial pilot has the burden of proof backwards

Medicare's AI prior-authorization pilot is being argued about as if the scandal were the algorithm or the bounty. I think the real defect is smaller, plainer, and more fixable than either: the program never has to prove a denial was correct, and the only party who can force the question is the patient. Flip that burden of proof — put the reviewer before the patient rather than after — and most of this fight evaporates.

Start with the money, because that's where the argument keeps getting stuck. As our brief on the released records laid out, CMS pays WISeR contractors 25 percent of the regional benchmark cost of each request they decline, with outside analyses putting the effective rate closer to 20 percent once discounts are counted. The agency's own Office of the Actuary wrote the consequence into a memo in as many words: participants "will have an incentive to deny as many claims as possible." On September 16, Sen. Patty Murray asked CMS deputy administrator Chris Klomp whether contractors make more money by denying care — "just 'yes' or 'no'" — and he answered, "My understanding is no." Murray's follow-up belongs in the record twice: "Okay, but they do deny care. They're paid to deny care."

I want to be careful here, because that exchange is emotionally satisfying and analytically weak on its own. A payment design is not a verdict on outcomes, and the people running the pilot are not villains for administering the program Congress and CMS handed them. The evidence that matters is what actually happened on the ground, and that part is genuinely ugly. Innovaccer asked CMS to delay the launch, then configured its system to auto-approve everything until it finished building the real one. Virtix denied more requests than it approved in its first quarter and was put on a corrective action plan after missing the program's 72-hour decision window. The roughly 1,000 pages the Electronic Frontier Foundation pried out of CMS through a FOIA suit it filed in March contain 5,944 denials across two vendors in three months, along with provider complaints about patients waiting for pain treatment. The GAO found in May that the administration set the program up without following proper procedure. CMS is planning to extend it to oncology, advanced imaging, pacemakers and genetic tests anyway.

Then there is the number that turns a bad launch into a structural problem. Only about 11 percent of Medicare Advantage denials get appealed, and roughly 80 percent of the appeals that do get filed succeed. If anything like that ratio carries over, most of these denials are wrong and almost nobody is checking. That is not an argument against prior authorization as such. It is an argument about where the reviewer sits. A program that pays for averted spending and audits almost nothing has effectively relocated the audit to the person in pain, on the assumption that they will be too exhausted to file a form.

There's a symmetry in this that the healthcare fight keeps covering up. The hospital side now bills with AI: the Blue Cross Blue Shield Association says AI coding tools added $942 million to its members' costs over two years — a figure the American Hospital Association calls unsubstantiated — while systems like McLaren Health Care buy chart-review software and report about a million dollars a month in added revenue. Insurers run AI to scan those same charts for reasons to deny. Both sides of the claims fight now own a machine that argues their case at volume, which is the subject of our coverage of the coding fight. The patient is the only participant without one — and the single channel built for them, the appeal, is the least-staffed mechanism in the system.

The strongest case for WISeR

The best defence of this program does not lean on the payment formula at all, and it deserves a fair hearing. Fee-for-service Medicare never required prior authorization for these services, and the procedures on the list — epidural steroid injections, cervical fusions and the rest of the roughly dozen — have well-documented histories of overuse and wild geographic variance. Every Medicare model that wants to slow spending reaches for shared savings, and paying a contractor a share of averted expenditure is a textbook version of it; the alternative is paying for volume, which is precisely what produced the $942 million on the other side of the ledger. CMS also says a qualified clinician reviews every denial, and about one in five appeals fails, which means the denials are not uniformly wrong. A 72-hour decision window, when met, would be faster than plenty of Medicare Advantage turnarounds patients already live with. Read charitably, this is an ordinary idea with an unusually bad launch.

Why the take holds anyway

The incentive was never the offence. The asymmetry of consequence is. A contractor that denies correctly earns 25 percent. A contractor that denies wrongly loses a share of that 25 percent only by scoring below 60 out of 100 — a penalty that trims profit without ever creating a loss. Denying is never the downside trade, which is exactly what the actuary meant. Compare the bounty model this borrows from: the False Claims Act works because the party bringing the case carries its own costs and faces sanctions for a bad one. That exposure is what makes a bounty enforcement instead of a revenue stream. WISeR took the bounty and left out the exposure.

Medicare already owns the instrument that fixes this, and it is not a better model. It is timing. Post-payment review — pay the claim, then audit the outlier vendors, claw back what doesn't survive scrutiny, and publish the reference standard — catches the same waste without conscripting patients as the audit layer. The technology claim underneath this whole argument is modest and worth stating plainly: nothing about a language model requires a denial to be issued before the evidence is examined. That ordering is a policy choice, not a capability limit, and the contracts show it was chosen.

What would change my mind

Three things, and they are all publishable by CMS within a quarter.

First, per-vendor accuracy: independent clinical review of a random sample of denials, with the wrongful-denial rate stated against published criteria. If that rate lands in the single digits, the harm here is paperwork and delay rather than wrongly refused care, and I've misjudged where the injury sits. Second, a symmetric penalty — claw back the bounty and add a fine on an overturned denial, and the actuary's line stops being a prediction and becomes a hypothesis nobody wants to test. Third, the boring paperwork test: every denial letter citing the criterion and the evidence behind it, a named clinician accountable for it, and abstention when the case is genuinely unclear instead of a default no. If those three arrive and the program's expansion goes ahead, I'll concede that the gate was worth having and the fix was administrative. Until then, the burden of proof is on the wrong desk, and the desk it's on belongs to someone who is already in pain.

Should the contractor or the government have to prove a denial was right? Tell us in the comments.

Sources: Electronic Frontier Foundation · EFF records: WISeR participation agreements · Ars Technica · STAT News · Sen. Patty Murray, HELP hearing · Blue Cross Blue Shield Association