California AB 1979: Health AI Must Keep a Licensed Clinician in Charge From 2027

California just drew a clear line around artificial intelligence in health care. A machine can help with a medical decision, but it cannot make that decision on its own. Governor Gavin Newsom signed Assembly Bill 1979 on September 30, 2026, and it takes effect on January 1, 2027. From that date, California hospitals, clinics, and doctors offices that use AI to help diagnose or treat patients must keep a licensed human in charge.

The bill, authored by Assemblymember Mia Bonta, covers clinical decision support systems: AI tools that predict, classify, or recommend things like a diagnosis, a treatment, or the timing of care. If the output shapes a patient’s care, a licensed clinician must use independent judgment. The software cannot perform clinical work reserved for licensed professionals, and it cannot direct unlicensed staff to do that work either. Scheduling tools and routine messages like appointment reminders are exempt.

What AB 1979 actually changes

The first change is about accountability. Providers must take reasonable steps to make sure a licensed clinician exercises independent judgment whenever AI output influences care. A doctor, nurse, or other licensed professional cannot simply pass along what the system suggests. They need to review it, weigh it against what they know about the patient, and own the decision. Reports on the law describe violations as unprofessional conduct, with enforcement possible through the Medical Board of California and other licensing boards.

The second change is about privacy. AB 1979 amends the Confidentiality of Medical Information Act, known as CMIA, to treat any business that offers a health care chatbot to consumers as a health care provider for CMIA purposes. A company running a health chatbot that handles medical information now carries the same confidentiality duties and penalties under CMIA that a provider would carry for that purpose. Analysts often describe CMIA as distinct from federal HIPAA rules, and in some respects more protective. A chatbot conversation about health is therefore not treated like an ordinary customer service chat.

The third change is about transparency. Health facilities and offices that use generative AI to generate written or spoken patient communications about clinical information must include a disclaimer that the communication was generated by AI, along with instructions on how a patient can contact a human provider. If a patient receives a message about their care, they should know when AI helped write it, and how to reach a person.

The wider package Newsom signed, and what he vetoed

AB 1979 did not arrive alone. Newsom acted on a small set of health AI bills before his September 30 deadline, signing three and vetoing two, according to a review by the California Health Care Foundation cited in coverage of the package. Alongside AB 1979, he signed Senate Bill 503, which requires companies that build clinical AI tools and facilities that use them to make reasonable efforts to spot systems that could give worse care to people based on traits such as race, sex, or age. He also signed Senate Bill 947, which limits how employers, including large health systems, can use automated software to discipline or fire workers, starting in July 2027.

The vetoes are just as telling. Newsom vetoed Senate Bill 903, a bill by Senator Steve Padilla that would have set strict rules on AI in psychotherapy, including a ban on advertising companion chatbots as therapy. He also vetoed Assembly Bill 2575, which would have protected clinicians from punishment for overriding AI recommendations. The result is a package that is firm on human oversight and privacy, but stops short of a therapy specific AI law. According to reports, the author of SB 903 plans to try again in a later session, so the therapy chatbot debate in California is paused, not settled.

Two days earlier, on September 28, 2026, Newsom also signed Assembly Bill 1609, which regulates customer service chatbots used by large businesses. It points the same way: people, not bots, must stay reachable and accountable.

Why this matters beyond California

California is the largest state market in the United States, and health technology companies rarely build one version of a product for California and a completely different version for everyone else. That is one reason the national conversation about AB 1979 has been lively since the October 29, 2026 Hunton Privacy Blog report highlighted the CMIA extension.

With most obligations taking effect on January 1, 2027, providers and AI developers have a short compliance window. Analysts advise organizations to start now, reviewing which tools count as clinical decision support, who reviews their output, and how chatbot data is stored and shared. For patients, the near term change may be subtle: a disclaimer on an AI generated message and a clearer path to a human. For clinics, workflows and documentation need to reflect that a licensed clinician, not the software, makes the clinical call.

It is also worth being clear about what AB 1979 does not do. It does not ban AI in health care, and it does not make chatbots illegal. It sets a floor. AI can support care, but it cannot replace the licensed professional at the center of care.

The message from Sacramento is clear: in California health care, the human stays in the loop, and the chatbot answers to privacy law.

Frequently Asked Questions

What is California AB 1979?

AB 1979 is a California law, authored by Assemblymember Mia Bonta and signed by Governor Gavin Newsom on September 30, 2026, that governs AI in clinical health care. It requires licensed clinicians to keep independent judgment when AI output shapes patient care, and it treats businesses offering health care chatbots as health care providers under the Confidentiality of Medical Information Act for CMIA purposes. It takes effect on January 1, 2027.

Does AB 1979 ban AI in hospitals and clinics?

No. The law does not ban AI in health care. It bars AI from independently performing clinical work reserved for licensed professionals, and it bars using AI to direct unlicensed staff to do that work. AI can still support diagnosis, treatment planning, and other clinical tasks, as long as a licensed clinician reviews the output and makes the final decision.

How does AB 1979 affect health chatbots?

Under AB 1979, a business that offers a health care chatbot to consumers is treated as a health care provider under CMIA, so it carries CMIA confidentiality duties and penalties for the medical information it handles. That means chatbot health conversations are covered by California medical privacy duties, rather than being treated as ordinary consumer chats.