How does AB 2575 change legal liability for AI developers when workers do not override AI?
Short answer
Assembly Bill 2575 changes how courts handle lawsuits involving harmful AI recommendations in health care. It prohibits AI developers and other responsible parties from arguing that a health care worker’s failure to override an AI system is a "superseding cause" of harm. This legal term refers to an event that breaks the chain of responsibility. By blocking this argument, the bill ensures that developers cannot shift full blame to the health care provider simply because the provider did not catch the AI’s mistake.
What the bill or law says
The proposed legislation adds Section 1714.48 to the California Civil Code. This section specifically addresses the liability of developers, modifiers, selectors, or deployers of clinical decision support systems (CDSS). The text states that these parties cannot claim in court that a health care worker’s failure to override the system’s output severs the defendant’s liability for alleged harm [https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB2575#94AMD].
However, the bill does not remove all legal defenses. Defendants may still present other affirmative defenses or provide evidence regarding comparative fault. This means that while they cannot use the specific "superseding cause" argument related to overriding the AI, they can still argue about other factors that might have contributed to the outcome.
How it works in practice
In a typical medical malpractice or product liability case, a defendant might argue that even if their product had a flaw, the final harm was caused by the professional who used it. Under current general legal principles, if a professional ignores a warning or fails to correct an obvious error, that action might sometimes relieve the manufacturer of responsibility.
AB 2575 alters this dynamic for AI in health care. If an AI tool provides a harmful recommendation and a doctor or nurse follows it without overriding it, the company behind the AI cannot tell the jury that the worker’s inaction is the sole reason for the injury. The developer remains accountable for the safety and efficacy of their product. This aims to keep the focus on whether the AI system itself was defective or dangerous, rather than solely on the human user’s reaction to it.
What the source does not answer
The provided bill text and status reports do not specify how courts will determine the percentage of fault between the developer and the health care worker in cases where comparative fault is allowed. The sources also do not detail how this liability standard interacts with existing federal regulations for medical devices. Additionally, the documents do not predict how insurance premiums for AI developers or health care facilities might change in response to this new liability framework.