California AI health care law sets new oversight bar for physicians

California practices have until Jan. 1, 2027, to make sure AI helps clinical care without replacing independent physician judgment. A new checklist from Harry Nelson and Nicholas Merkin outlines how health systems, clinics and telehealth operators can prepare for AB 1979.

Why it matters: - California is setting a new compliance test for AI in health care: licensed clinicians must keep the ability to exercise independent judgment when AI informs care. - Practices that rely on batch review or high-volume sign-off now face a risk that a signature alone will not satisfy the law. - The issue affects health facilities, clinics, physician offices and group practice offices.

What happened: - Harry Nelson and Nicholas Merkin published a checklist on AB 1979 ahead of the law’s Jan. 1, 2027 effective date. - Gov. Gavin Newsom signed AB 1979 on Sept. 30, after Assemblymember Mia Bonta (D-Oakland) carried the bill and the California Nurses Association sponsored it. - Newsom also signed SB 503, which targets bias in clinical decision support tools, and vetoed AB 2575 and SB 903. - Bonta said, “Technology should not replace human decision-making in health care.”

The details: - AB 1979 adds Section 22758.5 to the Business and Professions Code. - Covered practices must take reasonable steps to ensure that a licensed provider retains the ability to exercise independent professional judgment whenever care is informed by a clinical decision support system. - The law defines a clinical decision support system as an AI system that produces a prediction, classification, recommendation, evaluation or analysis to aid decisions about timing of care, diagnosis or treatment. - Practices may not use AI to independently perform any clinical function that the law reserves for a licensed person. - Practices also may not use AI to direct, guide, supervise or instruct unlicensed personnel in those reserved functions. - The law exempts documentation and communication tools that do not require professional judgment, including automated notices about record updates, appointment reminders and patient information-finding tools. - The law also excludes trainees using a tool in a supervised course of study. - Violations by physicians fall under the Medical Board of California or the Osteopathic Medical Board of California. - If a violation amounts to unlicensed practice, a licensing board may seek an injunction or restraining order. - AB 1979 allows AI-assisted decision support as long as professional judgment stays with licensed health care providers. - An analysis by the California Board of Behavioral Sciences says lawmakers removed earlier language that would have barred any clinical decision based solely on an AI system’s output. - Lawmakers also revised language that would have required a licensee to exercise independent judgment when reviewing and approving AI clinical decisions. - As enacted, the duty rests on the practice, which must take reasonable steps so the clinician retains the ability to exercise that judgment. - The law does not expressly set a minimum review time per patient or specifically address batch approvals. - The law does not automatically make existing AI-assisted workflows unlawful on Jan. 1, 2027. - AB 1979 also brings some consumer health chatbots under the Confidentiality of Medical Information Act. - A business offering a health care chatbot for managing consumer information, or for diagnosing, treating or managing a medical condition, is treated as a health care provider for CMIA purposes. - These requirements build on AB 3030, which has applied since Jan. 1, 2025. - When a practice uses generative AI to draft patient communications containing clinical information, the message must include an AI disclaimer and instructions for reaching a human, unless a licensed or certified provider reviews the message before it is sent.

Between the lines: - The checklist is aimed at a core gray area: when does human review become meaningful oversight instead of a rubber stamp? - High-volume telehealth workflows raise the clearest risk, especially when physicians are asked to approve many apparently low-risk cases at once. - The authors argue that a medical group cannot assume a technology vendor is solely responsible for compliance. - They also warn against reading the law so strictly that it blocks responsible screening and decision support, which could raise costs and slow care. - Training matters because physicians need to know when AI output should be questioned rather than accepted. - The authors point to their 2026 book, Digital Diagnosis: The AI Revolution in Healthcare and the Future of Regulation, as support for that concern.

What’s next: - Practices should sort administrative tools from tools that influence diagnosis, treatment or other licensed functions. - Practices should map how much information physicians see, how much time the workflow allows and whether rejecting recommendations is practical. - Practices should write rules for grouped review if physicians approve cases in batches. - Practices should document when physicians override AI recommendations and why. - Practices should train clinicians on the limits of each tool and when to disregard output. - Practices should verify that AI-drafted clinical messages carry the AB 3030 disclaimer or receive documented clinician review. - Practices should confirm that consumer chatbots comply with CMIA. - Regulators have not yet clarified how much physician involvement is enough, leaving records and workflow design as the main defense for now.

The bottom line: - AB 1979 does not ban AI in California health care. - It does force practices to prove that licensed clinicians still control clinical judgment when AI is part of the workflow.