Trump 政府 WISeR 项目用 AI 审批 Medicare 预授权，拒批率与激励结构引发争议
Key Highlights
Per Ars Technica, since January 2026 the Trump administration has piloted WISeR in six states, using AI to approve or deny prior authorizations for some Medicare services. In plain terms, what used to be decided by a human or insurance reviewer—"is this treatment approved"—now passes through an algorithm first, turning human judgment into an automated gate.
What Happened
Prior authorization is a U.S. healthcare gate: certain tests or treatments must be approved before reimbursement. WISeR uses AI for that gate, ostensibly to boost efficiency and cut manual work. But disputes arose: how denial rates are set, and whether the algorithm leans toward insurer savings over patient need.
Technical Details
The core risk in AI adjudication is "incentive misalignment"—if the optimization target is lower payouts, the model will naturally trend toward denial, while medical decisions should center on clinical need. Transparency, appeal mechanisms, and human-review fallbacks are indispensable guardrails, or a wrong denial could directly delay treatment.
Versus Competitors
Versus insurers' in-house review models, WISeR is government-led, with larger reach and policy weight. It turns "AI into public welfare" from debate into reality and pushes algorithmic accountability to the foreground, providing a precedent sample for other countries considering following.
Industry Impact
For the medical-AI industry, WISeR is a landmark "government buys AI decisions" case. It reminds practitioners that in life-and-power scenarios, models must be accurate and explainable, correctable, and supervised—or tech dividends become a public-trust crisis. In public-sector AI deployment, compliance and accountability are scrutinized before accuracy.
Why It Matters
The WISeR program using AI to approve or deny Medicare prior-authorization requests in six states puts an algorithm directly in front of patient care decisions. Replacing human or insurer reviewers with an automated gatekeeper changes who is accountable when care is delayed or denied.
The Stakes
The refusal rate and incentive structure are the flashpoints: if the AI is tuned to deny by default or rewarded for thrift, patients bear the cost in delayed treatment. The opacity of such systems makes it hard for beneficiaries to understand or contest decisions, which is a due-process concern.
Bottom Line
AI in benefit adjudication demands transparency and appeal rights by design. Efficiency gains are real, but they must not come at the expense of understandable, contestable decisions about people's health and coverage.
Looking Ahead
AI in benefit adjudication is likely to expand beyond the six-state pilot if it demonstrates cost savings, which makes the design of oversight, appeal rights, and transparency the deciding factor in whether it helps or harms beneficiaries. The efficiency case is strong; the safeguards are what determine fairness.
One More Angle
The deeper issue is accountability. When an algorithm denies care, someone must own the decision and the remedy. Building appeal paths and human review into the system from the start is not optional if the program is to survive scrutiny.
Closing Perspective
The WISeR program's use of AI to approve or deny Medicare prior-authorization requests is one of the clearest examples of generative AI moving from back-office assistant to front-line decision-maker in a high-stakes, heavily regulated domain. The efficiency argument is real and tempting: a vast backlog of prior-auth requests consumes enormous human effort, and an automated gatekeeper could in principle speed access to care and reduce administrative cost. The danger is equally real: if the system is tuned to deny by default or rewarded for thrift, patients pay in delayed or refused treatment, and the opacity of the model makes those decisions hard to understand or contest. Any deployment of this kind must therefore be judged less on its efficiency and more on its safeguards, including transparent criteria, robust appeal rights, and meaningful human oversight for adverse decisions. The six-state pilot is a test not just of the technology but of the policy framework around it. If the program expands without those guardrails, it risks becoming a cautionary tale about automating compassion out of a system that needs it, and regulators should treat the design of accountability as the central question.
In Short
Any deployment of this kind must be judged less on efficiency and more on safeguards, including transparent criteria, robust appeal rights, and meaningful human oversight, because automating compassion out of a system that needs it would make the program a cautionary tale.
Extended View
Using an algorithm to approve Medicare prior authorization tells a compelling efficiency story, but safeguards are what matter: transparent criteria, smooth appeal paths, and human review of adverse decisions. Without them, the program risks becoming a cautionary tale of automating care away.