Who Is Liable When Clinical AI Gets It Wrong?

A batch of letters published by STAT News crystallizes a question the healthcare industry has yet to answer: when an AI tool influences a medical decision and something goes wrong, who is responsible?

One reader put it bluntly, telling STAT that "a click, signature, or brief review should not magically transfer responsibility from an AI developer or platform to a physician." The concern is that vendors are building products designed to keep a human in the loop, then using that human's rubber-stamp approval as a legal shield. In practice, physicians are being asked to vouch for recommendations they cannot fully inspect or audit, while developers retain control over the underlying models.

The letters, which also touched on M.D. versus D.O. training, surrogacy, and other topics, reflect a broader unease as AI moves deeper into diagnosis and documentation. Until liability frameworks catch up, the risk falls unevenly on clinicians. Readers want accountability tied to whoever actually controls the system, not just the last person to click approve.