WASHINGTON, October 14, 2026 – California has decided that artificial intelligence can help doctors make decisions—but it cannot make them alone. Gov. Gavin Newsom signed Assembly Bill 1979 on Sept. 30, requiring health facilities to protect licensed clinicians’ independent judgment when AI-informed tools are used in patient care. The law takes effect Jan. 1, 2027.
At a hospital bedside, the distinction is practical. An algorithm may flag sepsis risk, recommend a treatment pathway, or analyze a patient’s deterioration. But a nurse who sees a patient worsening—or a physician who knows the chart does not match the model’s output—must be able to override it.
What the law requires
AB 1979, which was introduced by Assemblymember Mia Bonta from Oakland, includes hospitals, clinics, doctor’s offices, and group practices. It provides that the facilities must use all reasonable efforts to protect the ability of licensed providers to make independent professional judgments when care is guided by a clinical decision support system.
According to the act, clinical decision support systems include all kinds of artificial intelligence that are used to make predictions, classifications, recommendations, evaluations, and analyses about diagnosis, treatment, or the time for treatment. At the same time, the bill prohibits health facilities from using artificial intelligence to perform independently a clinical task that the law says has to be performed by a licensed individual, such as prescribing drugs or conducting testing.
Bonta said the goal is to prevent efficiency from becoming the only measure of good care. “Technology should not replace human decision-making in health care,” she said in a statement.
Nurses led the fight.
The California Nurses Association was behind the legislation. Its president, Sandy Reding, hailed the law as historic, saying that the nurses demanded that the AI technology “must aid, not replace” the clinical judgment of the nurses and other health-care professionals who take care of patients.
These issues were not theoretical in nature. According to News, the National Union of Healthcare Workers filed a complaint in July claiming that Kaiser Permanente had used an algorithm to evaluate patients seeking mental health care and come up with care-level recommendations without enough supervision by licensed clinicians.
Kaiser spokesperson Elissa Harrington said its AI and predictive tools “do not make care decisions,” adding that clinicians do. The health care system’s own research in 2020 showed that one deterioration alert tool helped cut deaths by 16 percent.
This is the dilemma of the whole discussion: AI can be beneficial if it works, but clinicians are afraid that it might become an invisible source of pressure to follow the machine.
Hospitals seek balance
California Hospital Association: Neutral.
Spokesman David Simon stated that the decision-making for clinical practice will remain within the domain of the physician while artificial intelligence would be able to assist in “thoughtful, measured” progress towards improvements.
California Medical Association, representing 50,000 doctors, also remained neutral despite previous statements being stricter. The reason is that the previous statement could limit the use of the tools of artificial intelligence.
Dr. Donald Hernandez, a physician dealing with hospitalized patients in Santa Cruz County, gave a realistic example. One of such tools can be characterized as “a fast librarian.” It is helpful but only if clinicians critically analyze the data.
Liability remains unsettled
The answer to the first question provided by AB 1979 is that the final decision regarding the clinical issue should be made by a licensed professional rather than by artificial intelligence software. The second question remains unanswered, however.
Healthcare attorney Henry Norwood expects more malpractice claims involving alleged overreliance on AI. “AI, we’ve seen, can be incorrect,” he said in an interview.
Newsom also rejected AB 2575, which would have prohibited health care facilities from using noncompliance with AI recommendations as a means of legal protection. Newsom argued that the law would not offer proper protections to clinicians and would limit regulatory bodies.
The message from California at this point is clear – AI can make medicine faster, but it cannot practice medicine. The question of liability for erroneous medical recommendations provided by the algorithm will be up next.
