Eliza Ward: Hey. Did you get through all of it?
Brian Reed: Most of it. Eight hundred pages is — I mean, I got through enough.
Eliza Ward: Eight hundred and something pages. The Electronic Frontier Foundation dropped them October 27th. All federal records. About WISeR.
Brian Reed: And the thing that hits me — this is a program that only launched January 1st of this year. Six states. Traditional Medicare. And we're already talking about corrective action letters going to at least four vendors within the first three months.
Eliza Ward: Right — and that's the part people need to sit with. Traditional Medicare, historically, almost no prior authorization. Like, basically none at scale. And then January 1st, 2026, that changes completely. AI contractors are now gatekeeping care in Oklahoma, Arizona, New Jersey, Ohio, Texas, Washington.
Brian Reed: So we went from essentially no prior auth to — this. In six months.
Eliza Ward: Yeah. And the EFF had to sue CMS to get these records. Not — wait, not a leak, not a voluntary disclosure. A FOIA lawsuit filed March 25th.
Brian Reed: The part I don't get is — what was CMS's reason for not releasing anything? What's the stated justification?
Eliza Ward: They didn't give one, really. And that — okay, that's actually less interesting to me than what the October 27th records actually show, because this isn't just CMS being cagey. The new thing, the thing that wasn't in September's thousand pages, is that CMS's own internal records acknowledge the payment structure. Like, the contracts themselves.
Brian Reed: Say that plainly. Because I think people are going to hear 'AI made bad decisions' and miss what's actually new.
Eliza Ward: The vendors get paid more when they deny more. That's in CMS's own records.
Brian Reed: Right — so imagine a parking lot attendant who gets a bonus for every car they turn away. You'd just... expect fewer cars getting in. That's not the attendant malfunctioning. That's the job they were given.
Eliza Ward: And KFF found targeted services dropped 15 to 19.5 percent in the first period of the pilot. That's not noise.
Brian Reed: Wait — 19.5 percent? In what, three months?
Eliza Ward: First period of the pilot. Yeah. And — hang on, because here's where I want to slow down — September's release was about operational chaos. Delays, broken portals, patient harm. Documented, bad, but fixable in theory. What October 27th adds is that the incentive structure is acknowledged internally. That's a different category of problem.
Brian Reed: So the question isn't whether the AI is calibrated wrong. It's whether you'd want it calibrated right if the contract still pays for denials.
Eliza Ward: Exactly — and that's what STAT News was pointing at in September when they reported the whole pilot was rushed. But the incentive piece? That's what EFF's second round actually surfaces. Four of six vendors flagged, corrective action letters sent, and we still don't know if those letters touched the payment structure at all.
Brian Reed: But that's the take everyone's skipping past — the 'AI failed' frame lets the transparency piece off the hook completely.
Brian Reed: Patients in Oklahoma, New Jersey, Ohio — they were already experiencing confusion, documented harm, long waits — and the public found out because EFF sued on March 25th. Not because CMS said anything. The harm came first. The disclosure came from a lawsuit.
Eliza Ward: Right — and those 66 vendor applications? Only surfaced through that same litigation. Someone picked these contractors. We have no idea on what criteria.
Brian Reed: Sixty-six companies applied. One process. No public record of how it was evaluated, and the selection criteria apparently weren't disclosed until — I mean, they're still not fully clear.
Eliza Ward: And EFF has said — this is their stated position — the public still doesn't have enough information to assess whether clinician review of AI recommendations is meaningful oversight or just... a signature on a denial.
Brian Reed: Nearly a year in and that's still an open question.
Eliza Ward: Nearly a year in. And — wait, this is what I want to press on — the wrong take circulating is that fixing the algorithm fixes the problem. It doesn't. If we don't know what human review actually looks like operationally, you could retrain the model tomorrow and the accountability gap stays exactly where it is.
Brian Reed: And the political piece is about to get loud — Dr. Oz is still defending this publicly, GAO just cleared Congress to use the Congressional Review Act against it. That tension is its own story, and we're going to get to it.
Eliza Ward: And that tension is not theoretical anymore — the GAO ruling is what makes this a different week. Because Dr. Oz is still out there defending WISeR publicly, the Trump administration is not pausing anything, CMS's response to flagged vendors was corrective action plans — standard regulatory step, not a suspension. But the Government Accountability Office just told Congress it has a procedural tool to actually block this.
Brian Reed: The Congressional Review Act. GAO cleared it.
Eliza Ward: Right. Which — I mean, that's not a press release. That's a formal procedural green light. Congress can move to nullify the WISeR model. Whether it does is a completely different question.
Brian Reed: And that's genuinely unresolved. Like, is there actually appetite for that? Because defending the CRA route requires someone to take political ownership of blocking an AI cost-cutting program while the administration is — I mean, Oz is framing this as fighting Medicare waste. That's not an easy vote.
Eliza Ward: No, it's not. And CMS issuing corrective action letters to four vendors doesn't change that math. Those letters are internal. Private. A beneficiary in, say, Ohio submitting a prior auth request right now — they have no way of knowing their vendor got a corrective letter, or what it said, or whether it touched the payment structure at all.
Brian Reed: Picture that person. November, somewhere in one of the six pilot states. Submitting a request to a vendor whose contract still rewards denial. And in Washington, the political argument is essentially — we sent the letter.
Eliza Ward: That's the accountability gap in concrete form.
Brian Reed: So the thing to watch is actually whether any member of Congress files a CRA resolution. That's the signal — not a hearing, not a statement. An actual resolution. Because the GAO ruling only matters if someone uses it.
Eliza Ward: And if nobody does — if it stays at the corrective-letters stage while Oz keeps defending it publicly — then the Center for Medicare Advocacy's concern, and EFF's, is that this just becomes the new baseline. Six states now. A six-year pilot. That's a long runway for a payment structure CMS's own records acknowledge is built around rejection.
Brian Reed: And that's — I mean, that's the split right there. If the diagnosis is 'the AI is bad,' you demand transparency requirements, maybe retrain the model. If the diagnosis is 'the payment structure is bad,' you have to restructure how vendors get paid. Those are completely different surgeries on completely different organs.
Eliza Ward: And right now the people running this program are saying neither organ is the problem.
Brian Reed: Because EFF's core unanswered question, nearly a year in, is whether meaningful human oversight of those AI recommendations actually exists inside WISeR. Not in theory. Operationally. And CMS hasn't answered it. So if Congress does file a CRA resolution and it goes nowhere, what does a beneficiary in one of those six states actually have? The corrective letters are private. The vendor criteria are proprietary. The human review layer is — we don't know what it is.
Eliza Ward: That's the open question. Not whether WISeR is efficient — whether the oversight is real. And the evidence doesn't answer it yet.
Brian Reed: Five more years on this pilot.
