NEWS

Can More Legally Defensible Clinical AI Increase Defensive Medicine?

September 7, 2026

More legally defensible clinical AI may come with an unexpected tradeoff: the systems that performed best against medical malpractice standards also recommended substantially more tests, procedures, and referrals.

A peer-reviewed study published in Communications Medicine evaluated 15 large language models using 198 U.S. medical malpractice cases in which courts had already established what appropriate care should have included. Researchers then compared how often each model recommended the court-endorsed action with the number and estimated cost of procedures it proposed.

The result raises a potentially important question for medical professional liability insurers and healthcare organizations: as AI tools become better at avoiding omissions that could later appear indefensible, could they also encourage more intensive—and potentially more defensive—patterns of care?

‍

What Did Researchers Find?

Across 3,072 simulated consultations, the models varied substantially in both legal defensibility and resource use. The highest-scoring model addressed the court-endorsed action in roughly 70% of consultations and recommended an average of 9.3 procedures per case. A lower-performing model reached the relevant standard in roughly one-third of consultations while recommending only 1.3 procedures.

Estimated Medicare procedure costs differed by more than fivefold between those models.

The relationship was remarkably consistent: models that scored higher for legal defensibility tended to recommend more medical interventions. The researchers found the same basic pattern when testing the models against a separate set of U.K. court cases.

‍

More Defensible Does Not Mean Proven Safer

The study did not evaluate physicians actually using these systems with patients, nor did it show that additional testing prevented malpractice claims or improved clinical outcomes.

That limitation matters. The researchers themselves identify several possible explanations for the pattern, including better clinical reasoning and a tendency by newer models to avoid omitting potentially relevant actions. The study therefore identifies a relationship worth watching rather than proving that AI will cause physicians to practice defensive medicine.

Even so, the liability implications are difficult to ignore. If clinicians increasingly rely on AI recommendations, questions may eventually arise not only when a physician ignores an appropriate recommendation, but also when following AI routinely generates tests, consultations, or procedures that may not otherwise have been ordered.

Our team is your team.

Get a quote

Physician Oversight Remains Central to the Liability Question

The emerging policy consensus continues to place clinical responsibility with humans rather than treating AI as an independent practitioner.

In June, the American Medical Association adopted new policies on clinical AI emphasizing that AI should support rather than replace physician judgment and that healthcare organizations need clear accountability, transparency, and oversight when these systems influence patient care.

That becomes particularly important when an AI recommendation conflicts with a physician's clinical judgment. Documentation of how the tool is used, who has authority to act on its recommendations, and how clinicians are expected to review AI output may increasingly become part of the risk-management picture.

‍

Clinical AI Is Becoming an MPL Underwriting Issue

For MPL underwriters, the relevant question is increasingly moving beyond whether a medical practice “uses AI.” How the technology is incorporated into care matters more. A tool used to summarize records presents a different exposure from one recommending diagnostic testing, triaging patients, or influencing treatment decisions. Hospitals and physician groups may therefore need to explain what systems they use, what clinical decisions those systems influence, and what safeguards exist around physician review.

For retail agents working with Western Summit, that distinction can become important when presenting technology-intensive practices to the market. AI adoption alone does not necessarily make a risk harder to place, but poorly defined responsibility and unclear clinical oversight can create underwriting uncertainty.

‍

The Liability Tradeoff Is Still Taking Shape

The new research does not establish that more capable clinical AI will increase malpractice exposure or healthcare costs in actual practice. It does, however, reveal a tension that deserves attention: recommendations that appear more defensible after a bad outcome may also encourage substantially greater intervention before that outcome occurs.

As healthcare organizations move clinical AI from experimentation into routine care, MPL carriers will increasingly need to understand not simply whether these tools work, but how they change physician decision-making. For agents and insureds alike, the quality of human oversight may prove as important as the quality of the algorithm itself.