An act to add Chapter 6 (commencing with Section 12898) to Part 2.5 of Division 3 of Title 2 of the Government Code, relating to automated decision systems.
SB 1248 would regulate how California state agencies use automated decision systems, including artificial intelligence, machine learning, statistical modeling, and data analytics, when those systems are used to help make decisions about public services. The bill defines “services” broadly to include public benefits, social services, employment assistance, vocational and education-related services, and the issuance, renewal, denial, or suspension of professional licenses or occupational credentials.
The bill would allow agencies to use automated systems to inform decisionmaking, but not to replace human judgment. It would prohibit a state agency from using an automated system’s output as the sole basis for an adverse action against a person, such as denying a benefit or license, unless another law expressly allows it. It also requires human review of outputs suggesting ineligibility or adverse action, disclosure when automated systems materially affect a decision, verification of output accuracy, periodic quality control review, and safeguards to reduce bias and protect personally identifiable, protected health, and other legally protected information. GovOps would be authorized to issue public guidance and provide technical assistance to agencies, with notice to the Joint Legislative Budget Committee before guidance is issued.
SB 1248 would add a new chapter to the Government Code governing state agency use of automated decision systems in service delivery and licensing decisions. It would not create a new appropriation, but it would impose operational requirements on state agencies, including human review, accuracy verification, nondiscrimination monitoring, privacy safeguards, and quality control processes. The bill would also expand GovOps’ role by authorizing it to publish guidance and provide technical assistance, while requiring legislative budget committee notice before guidance is released. Affected parties would include state agencies, licensing boards and bureaus, public benefit administrators, applicants for licenses or benefits, and vendors or third-party system providers used by agencies.
The bill appears generally favorable in committee and vote history, with unanimous or near-unanimous support in the recorded votes and no recorded opposition in the provided history. The bill’s findings frame automated decision systems as a modernization tool that could reduce delays in licensing and benefits administration while preserving fairness and accountability. Overall, the discussion reflected support for using technology to improve government efficiency, paired with a clear emphasis on safeguards.
The main points of contention are likely to center on how much discretion agencies retain, how burdensome the required human review and quality-control obligations will be, and whether the privacy and bias-mitigation requirements are sufficiently specific or too restrictive for practical deployment. Another likely issue is the bill’s scope, which covers both public benefits and professional licensing decisions, potentially affecting a wide range of agencies and programs. The bill’s supporters appear to prioritize faster processing and modernization, while concerns would likely come from those wary of algorithmic bias, due process risks, privacy exposure, and administrative costs associated with compliance.