An act to add Section 1714.48 to the Civil Code, to add Section 1339.76 to the Health and Safety Code, and to add Article 2.7 (commencing with Section 2820) to Chapter 2 of Division 3 of the Labor Code, relating to health care services.
AB 2575 would regulate the use of artificial intelligence and clinical decision support systems in California health care settings. It requires health facilities, clinics, physicians’ offices, and group practices that use these tools for patient care to give written notice to licensed health care professionals and other users about the tool’s developer, intended use, inputs, output methods, training and validation information, known risks and limitations, and ongoing maintenance. The notice must also state that direct-care workers may override an AI or clinical decision support output when, in their professional judgment, doing so is appropriate for the patient or necessary to meet the standard of care or comply with law.
The bill also creates a new Labor Code article protecting direct patient care workers from retaliation or discrimination for overriding, requesting to override, or relying in good faith on employer-approved technology. It declares state policy that workers must retain professional judgment in patient care and bars employers from using technology to replace or limit that judgment. Workers who believe they were retaliated against could file a complaint with the Labor Commissioner.
In addition, AB 2575 amends Civil Code Section 1714.48 to limit a liability defense in lawsuits involving AI or clinical decision support systems used in health care. A defendant that developed, modified, selected, or deployed such a system could not argue that a clinician’s failure to override the system was a superseding cause that breaks the chain of liability. The bill preserves other defenses, including causation, foreseeability, and comparative fault. It also states that violations of the new Health and Safety Code requirements may be enforced through existing licensing and unfair competition remedies.
The overall sentiment reflected in committee votes appears generally favorable but not unanimous. The bill advanced with majority support in multiple committees, including a 11-1 vote, a 5-2 vote, and a 9-4 vote, and it was later approved in committee 6-2 before being sent to Appropriations. That pattern suggests broad interest in AI oversight in health care, paired with some concern about the bill’s scope and regulatory burden.
The main points of contention appear to be the breadth of the disclosure requirements, the labor protections limiting employer control over clinical technology, and the liability provision affecting AI vendors and deployers. Supporters likely view the bill as preserving clinician judgment, transparency, and patient safety, while critics may worry it creates significant compliance obligations, expands potential liability, and could complicate adoption of AI tools in health care settings.
AB 2575 would add new disclosure, labor, and civil liability rules to California law for AI and clinical decision support systems used in patient care. It would amend the Civil Code, Health and Safety Code, and Labor Code, creating new obligations for health facilities, clinics, physicians, and group practices, and new worker protections enforced through the Labor Commissioner. It would also expose violations to existing health facility, medical board, and unfair competition enforcement mechanisms, while specifying that no state reimbursement is required for the bill’s mandates.
The bill appears to have received generally positive committee support, advancing through several committees with clear majority votes, though not unanimously. The vote margins suggest lawmakers broadly support increased oversight of AI in health care and protection of clinician judgment, but some members remain concerned about the bill’s regulatory and liability implications. Its referral to Appropriations indicates the fiscal and implementation issues were still being reviewed.
The most notable contention centers on how far the state should go in regulating AI-assisted clinical decisionmaking. Health care providers and employers may be concerned about the extensive notice and inventory requirements, the mandate to preserve worker override authority, and the prohibition on retaliation tied to technology use. AI developers and deployers may object to the liability rule that blocks a superseding-cause defense based on a clinician’s failure to override an output. At the same time, supporters likely argue these provisions are necessary to protect patient safety, prevent automation from displacing professional judgment, and ensure transparency about how AI is used in care.