An act to add Section 214.12 to the Revenue and Taxation Code, relating to taxation.
SB 420 establishes a new California framework for regulating “high-risk automated decision systems” used to assist or replace human decisionmaking in areas with significant consequences, such as employment, housing, lending, health care, education, essential utilities, legal services, and essential government services. The bill defines key terms, excludes low-risk or purely procedural tools, and applies primarily to developers and deployers of these systems in the state, with exemptions for small entities and certain federally approved systems.
The bill requires developers to conduct impact assessments before making covered systems publicly available, and requires deployers to conduct impact assessments within two years of deployment in most cases. Those assessments must describe the system’s purpose, intended outputs, data inputs, foreseeable impacts on protected classes, safeguards against algorithmic discrimination, monitoring practices, and how the system is being evaluated. For state agencies, the bill adds additional procurement rules and allows some agencies to opt out of their own assessment obligations if specified conditions are met.
SB 420 would add Chapter 24.6 to the Business and Professions Code and a new article to the Public Contract Code, creating new compliance duties for AI developers, deployers, and state agencies. It would require notice to affected individuals when a high-risk automated decision system is used, provide an appeal path to human review when technically feasible, mandate governance programs aligned with the NIST AI Risk Management Framework, and generally prohibit deployment of systems that are likely to cause algorithmic discrimination unless mitigated. It also authorizes enforcement by the Attorney General or Civil Rights Department, including civil penalties, injunctive relief, attorneys’ fees, and a 45-day cure process, while keeping certain impact assessments confidential and exempting trade secrets and privileged information.
The bill appears to have generally favorable support among lawmakers who voted on it, as reflected by multiple committee and floor votes with more yeas than nays, including a 26-9 Senate third-reading vote. The bill’s findings frame it as a consumer- and civil-rights-protection measure aimed at transparency, fairness, and accountability in AI-driven decisionmaking. At the same time, the repeated referrals to Appropriations and the suspense-file placement suggest concern about implementation costs and administrative burden.
The main points of contention are likely the breadth of the disclosure and assessment requirements, the potential compliance costs for developers and deployers, and the confidentiality of impact assessments. The bill tries to balance transparency with protection of trade secrets and legal privilege, but that tradeoff may concern both advocates for public access and industry stakeholders. Another likely issue is the bill’s restriction on deploying systems that are likely to produce algorithmic discrimination, which could be viewed as necessary consumer protection by supporters but as overly restrictive or difficult to operationalize by opponents. The small-business exemption and the state-agency procurement restrictions also suggest debate over how far the rules should reach and who should bear the compliance burden.