Medicare AI Pre-Approval Trial Brings Delays, Puzzling Denials and Painful Waits

01Six-State US AI Medicare Pre-Approval Trial Brings Reported Delays, Puzzling Denials and Painful Waits

Patients have cried in pain while waiting for treatment under a new Medicare review program, according to healthcare-provider feedback contained in federal documents released this month. Doctors also reported technical problems, long waits for decisions and denials they struggled to explain.

The complaints concern WISeR, short for Wasteful and Inappropriate Service Reduction, a Trump administration pilot launched in January. Medicare, the federal healthcare program for older Americans, had not previously required doctors to obtain advance approval for the services selected for the trial. WISeR placed those types of care under a process using artificial intelligence and machine learning to authorize or deny treatment.

The program operates in New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington. Officials say it is intended to ensure appropriate Medicare payments while reducing fraud, waste and abuse. The available source does not identify every service covered, quantify the number of affected patients or disclose how much authority AI has in any individual decision.

Reports of problems emerged soon after the rollout. They described technical difficulties, delayed decisions and treatment, puzzling denials and frustrated doctors, alongside patients left suffering while awaiting care. The Electronic Frontier Foundation, a digital-rights organization, released federal records obtained through litigation in September that largely confirmed those accounts. One provider called WISeR “a disgrace to the human race” and said patients had been seen crying in pain as they waited.

The pilot also faces a challenge over how it was created. In May, the US Government Accountability Office concluded that Trump administration officials had not followed the proper procedure when establishing WISeR, raising questions about the program’s legality. The source does not describe a court ruling that has halted it.

Lawmakers seeking more information have so far failed to force further disclosure. Last week, Representative Suzan DelBene, a Washington Democrat, called for a committee vote on releasing additional WISeR documents, but Republicans voted down the effort. DelBene accused the administration of concealing records and jeopardizing patients’ care.

Despite the provider reports, scrutiny of its creation and the unsuccessful congressional demand for documents, WISeR is continuing. The pilot is scheduled to run through the end of 2031, with expansion planned over the coming years.

02U.S. Appeals Court Upholds Pentagon’s Anthropic Supply-Chain Risk Designation, Delays Claude Ban Ruling

A federal appeals court in Washington, D.C., has upheld one of two Pentagon findings that labeled AI developer Anthropic a supply-chain risk. The designation bars the U.S. military from using Anthropic’s Claude models and prevents defense contractors from using them in work for the department.

Anthropic had been an early AI partner to several U.S. agencies and signed a $200 million Pentagon contract in July 2025. The relationship broke down during negotiations over adding Claude to the military’s GenAI.mil platform. The Pentagon wanted access for all lawful purposes, while Anthropic sought to exclude fully autonomous weapons and domestic mass surveillance. The Defense Department issued the risk designation in March.

In a 2-1 decision Friday, the D.C. Circuit panel rejected Anthropic’s claims that the ban was arbitrary, unauthorized and unconstitutional. Judge Gregory Katsas wrote that the Pentagon had ample support for concluding that Claude’s continued integration into department or contractor information systems presented a national-security risk covered by federal law.

Defense Secretary Pete Hegseth had raised concerns that an “overly constrained” model could shut down unexpectedly or be manipulated. Anthropic disputed those claims, but the majority said the ultimate responsibility for balancing such risks belongs to the president and defense secretary. Judge Neomi Rao joined the opinion, while Judge Karen LeCraft Henderson dissented.

The ruling does not mean Anthropic has finally lost the wider dispute. The Pentagon relied on two separate legal designations, requiring challenges in different courts. A federal judge in San Francisco ruled last month that the parallel designation was unlawful; Friday’s decision upheld the other one. The source does not establish how the two rulings will ultimately combine to determine the ban’s full scope.

The appeals panel delayed its decision from taking immediate effect so Anthropic can request another hearing from the same judges or seek review by the full D.C. Circuit. The company could also ask the Supreme Court to hear the case. Anthropic said it disagreed with the ruling and was considering its options, but has not identified which route it will take or the deadline before the decision takes effect.

03Sony and Universal sue Suno again, alleging v6 was trained on older-model outputs in “model laundering”

Sony and Universal Music Group have filed another lawsuit against Suno, alleging that the AI music company’s new v6 generator still benefits from copyrighted recordings used without permission. The labels, major holdouts that have not signed licensing agreements with Suno, say a new model cannot escape an older model’s allegedly infringing origins merely by learning from its outputs.

Suno has described v6 as a model trained from the ground up on a new dataset. At its launch, Suno’s Jack Brody said that dataset included “user data,” without detailing its contents. Spokesperson Rachel Racusen later told The Verge that Suno used material licensed from partners, community interactions including users’ creations and preference signals, and knowledge accumulated by its team.

The source does not establish whether the training data included audio uploaded by users, outputs generated from uploaded audio, or other outputs from Suno’s previous models. It also does not confirm that Suno used model distillation, a training method in which a new “student” model learns to reproduce the behavior or results of an older “teacher” model.

Those unresolved details sit at the center of the labels’ complaint. Sony and UMG allege that Suno’s older models were trained on unauthorized music obtained from YouTube and other sources, and that v6 was subsequently trained using user outputs produced by those models. The labels call that process “model laundering”: in their account, copyrighted expression passes from recordings into earlier models, then into their outputs, and finally into v6.

Sony further alleges that Suno used distillation to make v6 replicate results from its previous teacher models. That claim turns distillation into a new copyright battleground because v6 would not need to train directly on the labels’ recordings to inherit information the plaintiffs say was captured by its predecessors.

The complaint also argues that Suno retained unauthorized copies and that even a model not directly trained on Sony and UMG recordings could still benefit from them. These remain allegations, however: the source provides neither confirmation of the disputed training paths nor a court finding on the claims.