California has enacted a law requiring specified health care settings to preserve licensed clinicians’ independent professional judgment when artificial intelligence informs patient care. Assembly Bill 1979, signed by Gov. Gavin Newsom on Sept. 30, is scheduled to take effect Jan. 1, 2027, giving hospitals, clinics, physician offices and technology vendors time to assess how clinical AI tools are used.
The measure is being described in some coverage as a ban on AI making medical decisions independently. Its documented scope is narrower, though potentially consequential: it requires covered settings to take reasonable steps to keep licensed providers in control of professional decisions, and it restricts the use of AI to carry out clinical functions that state law reserves for licensed people. The law does not prohibit clinicians from using AI-supported predictions, recommendations or analyses in care.
What AB 1979 covers
Newsom’s announcement included AB 1979 in a package of California AI legislation intended to protect the role of doctors and other licensed providers when clinical decision tools are used in patient care. The bill was introduced by Assemblymember Mia Bonta.
Accounts of the law describe a clinical decision-support system as an AI system that produces a prediction, classification, recommendation, evaluation or analysis to aid choices about the timing of care, diagnosis or treatment. That could encompass a wide variety of software, from systems that flag a patient at risk of deterioration to tools that help prioritize diagnostic follow-up or recommend treatment options.
But the definition alone does not make all such software unlawful. Under reporting by Chief Healthcare Executive, the requirement is that a licensed provider acting within the provider’s scope of practice retain independent professional judgment when patient care is informed by a system. The law addresses the authority surrounding the tool, rather than barring software from generating clinical information.
It also bars covered facilities, clinics, physician offices and group practices from using or deploying an AI tool to independently perform a clinical function when that function must legally be performed by a licensed person. A related restriction concerns tools that direct, guide, supervise or instruct unlicensed personnel to perform functions reserved by law for licensed professionals.
Reported exceptions include AI used in supervised training toward licensure and automated decision tools that do not require the application of professional judgment. Those exceptions are important because health care has long relied on automation for some routine functions, such as calculations and administrative routing. The available accounts do not support treating AB 1979 as a blanket prohibition on clinical software or on every automated workflow in health care.
A human-oversight rule, not an outcome study
The law arrives as health systems and technology companies expand the use of generative AI and predictive tools in documentation, patient communication, scheduling and clinical decision support. Yet the legislation itself is a governance action, not evidence from a clinical trial. None of the available reporting establishes that AB 1979 will reduce medical errors, improve health outcomes or change malpractice claims after it becomes effective.
Its immediate practical question is organizational: whether health care entities can show that clinicians are able to exercise their own judgment when AI-generated material enters a care decision. Governing’s reporting quoted representatives of Kaiser Permanente and the California Hospital Association saying that clinicians, not AI systems, retain authority over patient-care decisions. A health care attorney cited by the publication said the law could also prompt malpractice allegations when a clinician’s reliance on AI guidance is questioned after an adverse outcome. That is a forecast of possible litigation, not an established effect of the new statute.
The phrase “reasonable steps” leaves room for interpretation. The public descriptions of AB 1979 do not lay out one mandatory workflow, documentation form or technical safeguard for every setting. Organizations may consider reviewing which tools are used in diagnosis or treatment decisions, clarifying who can override an AI output and training staff on the limits of a system. But those are potential compliance approaches, not requirements specifically identified in the available public accounts.
The law’s focus on tasks that require licensure also creates a boundary that may be clearer in some circumstances than others. A tool that calculates a score from patient data may be far removed from a system that effectively determines a diagnosis or directs an unlicensed employee to undertake a clinical act. How California applies the measure to particular products and workflows will depend on the underlying function, the role of the clinician and professional licensing rules.
Part of a broader California AI package
AB 1979 was signed alongside Senate Bill 503, a separate health-care AI measure. The governor’s announcement said the new protections require clinical-decision-tool developers to take reasonable steps to reduce known or predictable bias in the use of their systems. That provision addresses a different problem from AB 1979’s professional-judgment safeguard: bias mitigation concerns whether a system may work unevenly across patients or settings, while AB 1979 centers on who remains responsible for a clinical decision.
A Reed Smith analysis also says AB 1979 applies California medical-information confidentiality requirements to businesses offering consumer health-care chatbots. That extends the package beyond tools used inside hospitals and offices, though chatbot privacy rules and clinical-decision-support rules address distinct uses of AI.
For California health systems, the Jan. 1, 2027 effective date makes the law less an immediate change in bedside practice than a deadline for inventory and governance work. AI developers and the organizations that deploy their products will need to distinguish tools that merely support a clinician from systems that risk displacing professional judgment or performing functions the law assigns to licensed personnel. The statute preserves room for decision support, but places responsibility for patient care with the people licensed to provide it.