California lawmakers convene joint hearing on generative AI in health care; experts urge cautious, equity-focused rollout

Lawmakers and health leaders met in a joint informational hearing of the California State Assembly Health Committee and Privacy Committee to review how generative artificial intelligence is being used in health care and what state policy should require to protect patients and support equitable adoption.

The hearing gathered health-system executives, developers and patient advocates to describe early deployments — ambient scribes that draft clinical notes, AI that triages imaging and speeds treatment, and predictive models used in obstetrics — and to highlight persistent risks: bias in training data, unclear developer/deployer liability, privacy and data-access constraints, and the danger that under-resourced safety-net providers will be left behind.

“We are gonna talk about generative artificial intelligence in health care opportunities, challenges, and policy implications,” Assemblymember Rob Bonta said in opening remarks. Speakers emphasized potential benefits such as reduced clinician administrative time and faster identification of high-risk patients, while repeatedly urging robust oversight and monitoring.

Cedars-Sinai’s Craig Kwiatkowski described multiple internal projects, including ambient-scribe pilots and imaging workflows that rank scans for radiologists and notify response teams when the AI flags high-risk studies. Kwiatkowski said those workflows helped reduce time to mechanical thrombectomy by about 40 percent in a pulmonary embolism pathway and that ambient scribe pilots showed roughly a 20 percent reduction in time to notes and a 17 percent reduction in clinicians’ after-hours “pajama time.”

Dr. Daniel Yang of Kaiser Permanente described his organization’s large-scale rollout of an ambient-scribe product to clinicians and said the system has been used in millions of visits. “There’s only one way for me to guarantee that we can eliminate risks associated with AI…and that is for me to never deploy these AI technologies at all,” Yang said, arguing that foregoing AI would also mean keeping in place the current workforce and access problems he said the tools can help address.

Researchers and advocates pressed officials on data, equity and governance. Dr. Ziad Obermeyer (UC Berkeley) summarized prior research identifying algorithmic bias in health systems that used cost as a proxy for need: “Those tools were supposed to identify patients who were high risk…But instead of predicting who’s sick, those algorithms predicted who was gonna generate high health care costs,” he said, illustrating how biased training data can steer resources away from underserved patients.

Panelists and advocates recommended several state actions: clarify liability between developers and deployers, require standardized disclosures (model “fact labels” or model cards), invest in data infrastructure and workforce training for safety-net providers, and build monitoring and reporting requirements so deployed tools are validated locally and reevaluated over time.

The hearing did not produce votes or formal committee action. Chairs said the discussion would inform future policy work and legislative follow-up.

Despite differing views on speed of adoption, the assembled experts and patient advocates agreed on core themes: AI can improve efficiency and some clinical outcomes, but California will need clear rules and funding to ensure safety, preserve privacy, reduce bias and extend benefits to Medi-Cal and other underserved populations.