Health insurers are running prior authorization and post-service claims review through artificial intelligence. Plaintiffs in multiple lawsuits allege that carriers used those algorithms to deny claims in batches, without meaningful physician review, and two of those cases now carry names: Estate of Barrows v. Humana and Kisting-Leung v. Cigna. A Congressional Research Service report on AI health liability, published August 13, 2026, lays out what happens next. Almost all litigation over AI-related health injuries runs through state courts under state law, and no uniform national framework governs any of it.
Prior authorization is the step where a person decides whether a treatment gets paid for. That step is the loop, and the allegation is that the loop was cleared at machine speed. Courts are only now working out what an AI system even is in law. For years software was treated as a service, or shielded on First Amendment grounds, and no liability attached. More recently some courts have held that software counts as a product when it is proprietary in design, mass-marketed into the stream of commerce, and causes harm through design choices the developer controls. In Garcia v. Character Technologies a court applied exactly that reasoning to a chatbot, and found Google could face liability as a component part manufacturer because its language model was built into the thing that caused the harm.
Whether a denied patient has any recourse at all can come down to which kind of plan they happen to have. Every state has an unfair and deceptive practices statute, but courts have found the Medicare statute preempts those claims against Medicare Advantage plans, while a district court found that ERISA did not preempt the same claims against private insurers. States are legislating into the gap one bill at a time, Arizona House Bill 2175 covers AI in coverage determinations, and in Congress the Senior Chatbot Protection Act would regulate general-use chatbots aimed at older Americans while leaving state protections standing. The White House is moving the other way. An executive order directs the Attorney General to stand up an AI Litigation Task Force to challenge state AI laws that conflict with a minimally burdensome national policy framework.
No court has ruled on the merits in any of these cases yet, and that is itself the finding. The harm surfaces through lawsuits filed long after the denial letter arrives, not through anything inside the insurer that catches it while it is happening, which is how a fight with no verdict still scores 15 out of 100: three for oversight, four for monitoring, four for incident response, four for accountability. Take the doctor out of the review to clear the queue faster and the human in the loop does not disappear, the patient becomes the only one left in it.
