Issue 012 looked at AI in the tools you use inside your own practice. This issue looks at the other side of the table. In June the American Medical Association adopted policy against AI making coverage and prior authorization decisions without a physician in the loop. In the same month Congress moved to defund the federal government’s own AI prior authorization pilot in Medicare. And a peer-reviewed analysis traced the real problem upstream, to the value choices a tool is configured to make before anyone uses it. Three bodies, working independently, pointed at the same seat. When AI decides whether a patient gets care, who is allowed to overrule it, and can they see how the decision was made.
Signal 1
The AMA Adopted Policy Opposing Autonomous AI in Coverage and Prior Authorization Decisions, and Calling for Annual Audits of the Tools (June 2026)
At its annual meeting in Chicago, held June 5 to 10, 2026, the American Medical Association House of Delegates adopted new policy on the use of AI in coverage and utilization management. The policy opposes the use of autonomous or semiautonomous AI systems as a substitute for physician review in coverage determinations, and it calls for regulations that require these tools to operate inside physician-led processes rather than around them. It asks for greater transparency when AI is used in prior authorization and other utilization decisions, including disclosure of the clinical logic, data sources, and guidelines used to reach an adverse determination. It also calls for regular audits of AI-driven clinical review tools, with reaudits triggered by material changes to the model, its training data, or the applicable clinical guidelines, and a comprehensive audit at least once a year regardless of those changes. The AMA continues to use the term augmented intelligence to signal that the technology is meant to assist a physician, not stand in for one. Sources: American Medical Association, June 2026; Healio, June 11, 2026; TechTarget Healthtech Analytics, June 2026; Fierce Healthcare, June 2026.
What this means for you
The contested ground has moved from the tool in your office to the tool a payer uses to deny your patient. The AMA is putting on record that an adverse AI determination should arrive with its reasoning attached and a physician standing behind it. That gives you something concrete to ask for the next time a denial lands. It also tells you where the standard is heading, because the body that defines professional expectations has now written down that a denial engine running without human review is not acceptable practice. The same logic reaches the AI you bring into your own work, where the question of who reviews the output before it touches a patient is identical.
Signal 2
Congress Moved to Defund WISeR, the CMS Pilot Putting AI Into Prior Authorization for Traditional Medicare (June 2026)
The Wasteful and Inappropriate Service Reduction model, known as WISeR, began on January 1, 2026, and runs through 2031 in six states: New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. It pairs AI and machine learning with human clinical review to screen selected items and services in traditional Medicare for prior authorization, and CMS states that any recommendation for non-payment is made by a licensed clinician applying existing coverage policy. In June 2026 the House Appropriations Committee approved language that would bar federal funds from being used to implement the model. The committee action reflected bipartisan concern over transparency, bias, accountability, and the effect of AI-assisted review on patient access to care. CMS has maintained that WISeR does not change Medicare coverage or payment policy and that the model only applies established rules. Sources: CMS WISeR Model materials, 2026; Medical Economics, 2026; National Health Law Program, 2026; Telehealth.org, 2026.
What this means for you
Even the federal government’s own pilot of AI in prior authorization is being contested by the body that funds it. The pattern worth your attention is the requirement that keeps surfacing on both sides of the argument. CMS defends WISeR by stressing that a licensed clinician owns every denial, and the critics attack it on whether that review is real and visible. Whether the AI sits inside a private payer or inside Medicare, the defensible position both the AMA and Congress are converging on is the same. The no has to be reviewable, and a person has to answer for it. That is the line you can hold a payer to today.
Signal 3
A Peer-Reviewed Analysis Argued That AI in Clinical Workflows Quietly Encodes Value Tradeoffs, and That Accountability Belongs at Procurement, Not Only at Model Design (March 30, 2026)
Writing in npj Digital Medicine on March 30, 2026, Shefali Patil, Christopher Myers, and Tinglong Dai argued that adaptive AI systems embedded in clinical decision-making operationalize hidden value tradeoffs at scale. When a tool weighs cost against benefit, or speed against thoroughness, it is making a value choice, and that choice is often buried in a configuration the clinician never sees. The authors describe how this constrains a clinician’s ability to weigh competing priorities for a given patient, and how it shifts new liability pressure onto the people using the tool rather than the ones who built it. Their proposed remedy moves attention away from the technical design of the model and toward procurement and implementation accountability: standardized transparency through model cards, paired with internal, multidisciplinary institutional review to deliberate and document the value-laden settings before a tool goes live. Sources: Patil, Myers, and Dai, npj Digital Medicine, volume 9, article 269, March 30, 2026.
What this means for you
The decision a coverage AI makes is not neutral, and neither is the one a clinical tool makes in your office. Someone chose the thresholds. The practical implication is that the most important governance work happens before the tool is ever used, at the moment you decide to bring it in and on what terms. If you cannot say what tradeoffs a tool was configured to make, you have not finished evaluating it. The AMA policy and the Congressional move are both downstream attempts to repair decisions that should have been examined at procurement, which is the one point in the process where you still have the standing to ask the question and walk away.
The Pattern
Three bodies, working independently, pointed at the same seat. The AMA wrote policy against AI making coverage and prior authorization decisions without a physician in the loop. Congress moved to defund the federal government’s own AI prior authorization pilot. And a peer-reviewed analysis located the real problem upstream, in the value choices a tool is configured to make before anyone uses it. Read together, 2026 is not a debate about whether AI belongs in health care. It is a narrower fight. When AI decides whether a patient gets care, who is allowed to overrule it, and can they see how the decision was reached. The answer taking shape is that a licensed human has to own the denial and the reasoning behind it has to be visible. That standard does not stop at the payer’s door. It is the same standard that applies to any tool you bring into your own practice.
One Thing You Can Do This Week
The next time an AI-assisted denial crosses your desk, from a private payer or a Medicare contractor, ask for two things in writing. First, the clinical logic and guidelines the determination relied on. Second, confirmation that a licensed clinician reviewed it. Both are now things the AMA is on record asking be made available. If you receive them, you can answer the actual reasoning instead of the outcome. If you cannot get them, you have documented the exact gap the people writing the rules this year are trying to close, and that record is worth keeping.