September 24, 2026 by matray

California legislation that would establish new liability rules for artificial intelligence used in patient care is awaiting action by Gov. Gavin Newsom, who has until Sept. 30 to sign or veto the measure.

Assembly Bill 2575, which cleared the Legislature Aug. 31, and was presented to Newsom Sept. 15, would prevent AI developers and deployers from escaping liability by arguing that a healthcare professional should have overridden an AI-generated recommendation.

The legislation would establish protections for healthcare workers using clinical decision-support systems while addressing how responsibility may be allocated when an AI system contributes to patient harm.

Under the bill, a defendant that developed, modified, selected or deployed a clinical decision-support system could not argue that a healthcare professional's failure to override the system's output was a superseding cause that severed the defendant's liability for the alleged harm.

The provision could have significant implications for medical professional liability claims involving AI.

For example, if an AI system recommends against additional testing and a physician follows that recommendation, the developer or organization responsible for deploying the system could not avoid liability simply by arguing that the physician should have recognized the erroneous recommendation and overridden it.

The legislation would not, however, immunize physicians or other healthcare professionals from malpractice liability. Clinicians would remain responsible for meeting applicable professional standards of care, and defendants could still present evidence concerning the fault of other parties, causation and foreseeability.

The result could be claims in which responsibility for an adverse outcome is divided among physicians, healthcare organizations and AI developers or vendors rather than placed solely on the clinician who made the ultimate treatment decision.

AB 2575 also seeks to preserve clinicians' independent professional judgment when using AI. It declares it California public policy that workers providing direct patient care remain free to exercise professional judgment when making patient-care decisions.

Employers generally would be prohibited from retaliating or discriminating against healthcare workers based solely on their decision to override or rely upon the output of a clinical decision-support system. Those protections would not determine whether the clinician satisfied the standard of care in a professional negligence or malpractice action.

The legislation also would impose transparency requirements on healthcare organizations using clinical decision-support systems. Healthcare facilities, clinics, physician offices and group practices would be required to make information about the systems available to healthcare professionals and other workers who use or review their outputs.

The measure comes as hospitals and medical practices increasingly incorporate AI into clinical decision-making, raising questions for liability insurers about how responsibility should be allocated when technology influences a diagnosis, treatment recommendation or other clinical decision.

An Assembly analysis of AB 2575 noted that an estimated 65% of U.S. hospitals already use AI tools, including systems used for clinical decision support, patient monitoring, electronic health records, staffing and predicting patient outcomes.

The legislation has also generated opposition from healthcare organizations. The California Hospital Association and other healthcare stakeholders have argued that the proposal could create problematic liability incentives, interfere with quality oversight and discourage adoption of beneficial AI technology. The association continues to list its position as opposed to AB 2575 while noting that the bill is pending action by the governor.

Supporters, including the California Nurses Association, argue that the measure would protect healthcare workers' ability to exercise independent professional judgment and prevent AI developers and deployers from shifting responsibility for harmful AI outputs onto clinicians.