---
id: "lib-2026-10-27-new-records-obtained-by-eff-expose-more-problems"
type: "library"
title: "New records obtained by EFF expose more problems with Medicare's AI experiment"
lang: "EN"
org: "EFF"
source_type: "Report"
date: "2026-10-27"
week: "2026-W44"
url: "https://nosov.ai/library/lib-2026-10-27-new-records-obtained-by-eff-expose-more-problems/"
original_url: "https://www.eff.org/deeplinks/2026/10/new-records-obtained-eff-expose-more-problems-medicares-ai-experiment"
tier: "B"
tier_name: "Useful, limited"
tier_reason: "EFF, an advocacy group in litigation with the government, publishes primary records from a Medicare AI prior-authorisation pilot; the documents are real evidence, but the selection and framing are argued, and no outcome rates are given."
summary_basis: "full-text"
original_minutes: "8"
tags: ["ai technology", "future of work", "enablement", "business transformation", "public sector ai", "prior authorisation", "oversight"]
---

# New records obtained by EFF expose more problems with Medicare's AI experiment

Tier B (Useful, limited): EFF, an advocacy group in litigation with the government, publishes primary records from a Medicare AI prior-authorisation pilot; the documents are real evidence, but the selection and framing are argued, and no outcome rates are given.

Summary based on the full text.

## In short

An Electronic Frontier Foundation post (27 October 2026) releasing records on WISeR, a US Medicare pilot in which private vendors use AI to evaluate prior-authorisation requests for some services. In its first three months, the agency sent formal requests about delays, technical failures and possibly inaccurate denials to at least four of the six vendors, without disclosing this publicly. EFF argues that human review of AI denials is a weak safeguard and asks Congress for legal limits.

## Key numbers

- 800+: pages of new records released, obtained through EFF's lawsuit
- 4 of 6: WISeR vendors that received information requests or corrective-action requests in the first three months
- 66: companies that applied to the WISeR model, 6 selected
- six months: time from announcement to launch, which the agency itself called an extremely condensed timeline

## What it reports

- Records include agency letters to four of six vendors and their responses, the applications of 66 companies, and contracts for evaluating and monitoring the pilot.
- One vendor was asked about turnaround times beyond guidelines, denials "inconsistent with applicable criteria" and poor customer service; another about late determination letters; a third received two corrective-action requests and a non-compliance notice over a missing provider portal.
- Vendors' responses cite the compressed timeline; one says it warned the agency months ahead that its portal might not be ready.
- Information a vendor provided on its automation safeguards appears redacted in the released records.
- The payment structure gives vendors, in the government's own words, "an incentive to deny as many claims as possible"; one vendor denied more requests than it approved in the first months.
- Cited elsewhere: a system at another insurer let reviewers reject claims without opening files, averaging 1.2 seconds per case.

## How they got it

Records obtained through a Freedom of Information lawsuit filed in March 2026, published with excerpts from agency and vendor letters. Earlier releases covered about 1,000 pages.

## Limits

- Advocacy source with a stated position against AI in benefit decisions.
- No denial, error or appeal rates for the pilot; harm is inferred from correspondence.
- Corrective-action requests show problems were flagged, not how serious they were or whether fixes worked.
- The 1.2-second figure refers to a different company and program.

## Why it matters for companies

Any firm automating approval or denial decisions, in insurance, credit, claims or HR, faces the same questions the records raise: whether reviewers have time and authority to override, what data drives individual decisions, and whether incentives reward denial. A "human in the loop" clause is not evidence of oversight unless review time, override rates and data inputs are measured and can be shown to a regulator.

Original: https://www.eff.org/deeplinks/2026/10/new-records-obtained-eff-expose-more-problems-medicares-ai-experiment