<h2>At a glance</h2> <p>Assembly Bill 2575 (AB 2575), introduced by Assembly Member Liz Ortega, regulates the use of clinical decision support systems (CDSS) and artificial intelligence in California health care settings. The bill aims to increase transparency regarding how these technologies function and protect the professional autonomy of health care workers. If enacted, the measure would require health facilities to provide staff with detailed information about the AI tools used in patient care upon request and prohibit employers from retaliating against workers who override AI recommendations based on their professional judgment.</p> <p>The bill has passed the State Assembly and is currently under review in the Senate. As of August 3, 2026, it was referred to the Senate Appropriations Committee’s suspense file, indicating that lawmakers are reviewing its potential fiscal impact before proceeding. For more context on how California voters and residents engage with policy changes, visit the <a href="/california/">California voter initiatives and community proposals</a> homepage.</p> <h2>What the bill would do</h2> <p>AB 2575 introduces new requirements across three areas of state law: the Civil Code, the Health and Safety Code, and the Labor Code. The central focus is on "clinical decision support systems," which the bill defines as automated systems or generative artificial intelligence that produce predictions, classifications, or recommendations used to inform clinical decision-making for patient care.</p> <h3>Transparency and Inventory Requirements</h3> <p>Under the proposed Health and Safety Code Section 1339.76, health facilities, clinics, physician offices, and group practices that use CDSS for patient care must maintain an inventory of these systems. By July 1, 2027, these entities must make this inventory available to licensed health care professionals and other personnel who use or view outputs from these systems upon request. The inventory must be updated at least annually.</p> <p>The bill specifies that the information provided must include:</p> <ul> <li>A summary of the system’s developer and a description of the output it produces.</li> <li>The intended use of the system, including the target patient population and its role in supporting clinical decisions.</li> <li>Cautioned out-of-scope uses, including known risks and limitations.</li> <li>A summary of how the system generates outputs.</li> <li>A summary of the training data or clinical research underlying the recommendations, including demographic representativeness and known biases based on protected characteristics.</li> <li>A summary of the validation process and qualitative measures of performance.</li> <li>A link to the Certified Health IT Product List produced by the federal Office of the National Coordinator for Health Information Technology.</li> </ul> <p>Facilities must also notify relevant staff members upon hire and annually of their right to request this inventory. These requirements do not apply to AI used solely for documentation, communication, or administrative tasks that do not involve professional judgment, such as automated messages about health records.</p> <h3>Protection of Professional Judgment</h3> <p>The bill adds Article 2.7 to the Labor Code, declaring it state policy that workers providing direct patient care must be free to use their professional judgment within their scope of practice. It explicitly prohibits employers from retaliating or discriminating against workers who override or rely on CDSS outputs when making assessments or decisions. Workers who face retaliation for exercising this judgment may file a complaint with the Labor Commissioner.</p> <h3>Liability Standards</h3> <p>In the Civil Code, the bill adds Section 1714.48, which addresses legal liability. It prohibits developers, modifiers, selectors, or deployers of a CDSS from arguing in court that a health care worker’s failure to override the system’s output is a "superseding cause" that severs the defendant’s liability for alleged harm. This means that if an AI system causes harm, the company behind it cannot blame the health care worker for not catching the error as a way to avoid responsibility. However, defendants may still present other affirmative defenses or evidence regarding comparative fault.</p> <h2>Why supporters may favor it</h2> <p>Proponents argue that the bill enhances patient safety by ensuring that health care providers have access to critical information about the tools they use. By requiring disclosures about training data, biases, and limitations, the bill aims to prevent blind reliance on technology that may not be suitable for all patient populations. Supporters also emphasize the importance of protecting the professional autonomy of doctors, nurses, and other health care workers. They contend that legal protections against retaliation are necessary to ensure that human expertise remains the final authority in patient care, rather than algorithmic outputs. Additionally, clarifying liability standards ensures that technology developers remain accountable for the safety and efficacy of their products.</p> <h2>Why critics may object</h2> <p>Critics may raise concerns about the administrative and financial burdens placed on health care facilities. Maintaining detailed inventories and providing technical summaries of AI systems could require significant resources, particularly for smaller clinics and physician offices. Some may argue that restricting legal defenses for AI developers could increase litigation risks, potentially discouraging innovation or the adoption of beneficial clinical decision support tools. There may also be questions about the practicality of enforcing these provisions and the clarity of what constitutes sufficient disclosure of complex algorithmic processes.</p> <h2>Who may be affected</h2> <p>The primary groups affected by AB 2575 include:</p> <ul> <li><strong>Licensed health care professionals:</strong> Doctors, nurses, and other clinical staff who use AI tools in their daily work will gain new rights to access information about these tools and protection from retaliation when overriding them.</li> <li><strong>Health care facilities:</strong> Hospitals, clinics, and physician offices will need to implement new compliance procedures to maintain inventories and provide notifications to staff.</li> <li><strong>AI developers and deployers:</strong> Companies that create or sell clinical decision support systems will face new liability standards that prevent them from shifting blame to health care workers for failures to override AI outputs.</li> <li><strong>Patients:</strong> Individuals receiving care may benefit from increased transparency and the assurance that their providers are using AI tools with full awareness of their limitations and biases.</li> </ul> <h2>What the vote record shows</h2> <p>AB 2575 has seen mixed but generally supportive voting patterns in the Legislature. In the Assembly, the bill passed its third reading on May 27, 2026, with 48 votes in favor and 15 against. Prior to that, it received favorable recommendations from several committees, including the Committee on Health and the Committee on Labor and Employment. In the Senate, the bill has been amended multiple times and passed out of committee reviews, including the Committee on Health and the Committee on Labor, Public Employment, and Retirement. However, its progress halted temporarily when it was referred to the Senate Appropriations Committee’s suspense file on August 3, 2026. This step is common for bills with potential fiscal impacts, as the committee evaluates whether the costs associated with the measure are manageable within the state budget.</p> <h2>What happens next</h2> <p>Because AB 2575 is currently in the Senate Appropriations Committee’s suspense file, its immediate future depends on the committee’s decision to release it for further consideration. If released, the bill would proceed to the Senate floor for a vote. If it passes the Senate, it would then go to the Governor for signature or veto. Given the complexity of regulating artificial intelligence and the potential costs for health care providers, lawmakers may continue to refine the language to balance safety, accountability, and operational feasibility. Readers interested in other health care legislation can explore related stories, such as <a href="/blog/california/ab-1930-california-shield-out-of-state-health-care-investigations">AB 1930: California’s Proposed Shield Against Out-of-State Health Care Investigations</a>, which addresses different aspects of health care regulation and privacy.</p> <h2>Sources</h2> <ul> <li><a href="https://leginfo.legislature.ca.gov/faces/billStatusClient.xhtml?bill_id=202520260AB2575">Official Bill Status Page</a></li> <li><a href="https://legiscan.com/CA/bill/AB2575/2025">LegiScan Record for AB 2575</a></li> <li><a href="https://legiscan.com/CA/text/AB2575/id/3449525">Amended Bill Text (June 18, 2026)</a></li> <li><a href="https://leginfo.legislature.ca.gov/faces/billVotesClient.xhtml?bill_id=202520260AB2575">Roll Call Votes</a></li> </ul>

Keep exploring California civic power

Explore current California voter initiatives and community proposals. Browse the complete California civic reporting library.

Related California reading

Sources