WHAT THIS BILL REGULATES · 3 REQUIREMENT TYPES
How Is This Bill Enforced
Verbatim statutory text on the left; plain-language analysis and a per-section checklist on the right. Numbered markers cross-link to the matching checklist row.
(a)–(f) The Legislature finds and declares all of the following: (a) A multistakeholder regulatory organization (MRO)Multistakeholder regulatory organization (MRO)"Multistakeholder regulatory organization (MRO)" means an entity designated as an MRO by the Attorney General pursuant to this chapter that performs the functions specified in Section 8898.3, including certification of developers' exercise of heightened care and compliance with standards based on best practices for the prevention of personal injury and property damage with respect to an artificial intelligence model or application.Gov. Code § 8898(d) tasked with defining standards based on best practices and certifying adherence to them is an agile, public-private model designed to promote innovation, ensure the security of artificial intelligence (AI) platforms, reduce regulatory uncertainty, and build societal trust. (b) By proactively setting clear standards, creating tailored pathways for both established companies and emerging developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c), and offering legal and economic incentives, the MRO transforms compliance into a competitive advantage. It is not just about managing risks; it is about accelerating responsible growth and empowering businesses to confidently innovate and thrive in an AI-driven economy. Compliance with established standards confers a strong market advantage. (c) Leveraging private sector and government cooperation to achieve what would otherwise require regulations is a proven approach that utilizes all available expertise while enhancing transparency among industry players, policymakers, and the public. (d) Rather than relying on government agencies, semiprivate standards organizations with sector-specific expertise can better accommodate diverse market participants, varied technology use cases, and aligned public-private interests. This approach creates an adaptable and predictable compliance mechanism that ensures AI governance standards can evolve quickly alongside technological advancements. (e) Legal safeguards are effective means to incentivize responsible AI development and prevent unnecessary harm. Reducing litigation risk encourages investment and fosters innovation. It also promotes heightened care and adherence to best practices while maintaining accountability and effectively balancing risk mitigation with consumer and public protection. (f) Public opinion research shows that while the public wants government to help establish guardrails for AI, a majority believe the government alone is incapable of effectively establishing these guardrails. An MRO surpasses traditional regulation by incentivizing a race to the top for transparency and safety that prioritizes innovation and adaptability and serves as a central, informed voice to governments and society for responsive, forward-looking governance.
Section 1 sets forth legislative findings supporting the creation of multistakeholder regulatory organizations as a public-private model for AI safety governance. The findings emphasize that MROs can leverage private-sector expertise, promote innovation, reduce regulatory uncertainty, and incentivize a 'race to the top' for transparency and safety. The Legislature notes that public opinion supports government guardrails for AI but doubts government's capacity to establish them alone.
This section creates no compliance obligations and is purely declaratory.
(a)–(f) As used in this chapter: (a) "Artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a)" means a software program or system that uses artificial intelligence modelsArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) to perform tasks that typically require human intelligence. (b) "Artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b)" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments. (c) "DeveloperDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c)" means a person who develops an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) that is deployed in the state. (d) "Multistakeholder regulatory organization (MRO)Multistakeholder regulatory organization (MRO)"Multistakeholder regulatory organization (MRO)" means an entity designated as an MRO by the Attorney General pursuant to this chapter that performs the functions specified in Section 8898.3, including certification of developers' exercise of heightened care and compliance with standards based on best practices for the prevention of personal injury and property damage with respect to an artificial intelligence model or application.Gov. Code § 8898(d)" means an entity designated as an MRO by the Attorney General pursuant to this chapter that performs the functions specified in Section 8898.3, including certification of developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c)' exercise of heightened care and compliance with standards based on best practices for the prevention of personal injury and property damage with respect to an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or application. (e) "PlanPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e)" means a planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) submitted pursuant to Section 8898.2. (f) "Security vendorSecurity vendor"Security vendor" means a third-party entity engaged by an MRO or developer to evaluate the safety and security of an artificial intelligence model or application through processes that include red teaming, risk detection, and risk mitigation.Gov. Code § 8898(f)" means a third-party entity engaged by an MRO or developerDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) to evaluate the safety and security of an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or application through processes that include red teaming, risk detection, and risk mitigation.
Section 8898 defines the key terms used throughout the chapter, including artificial intelligence application, artificial intelligence model, developer, multistakeholder regulatory organization (MRO), plan, and security vendor. The AI model definition is notably broad — any engineered or machine-based system that infers from inputs how to generate outputs influencing physical or virtual environments — and is not limited by compute thresholds or capability levels.
(a)–(b) 1 The Attorney General shall designate one or more MROs pursuant to this chapter. (b) In complying with subdivision (a), the Attorney General shall determine whether an applicant MRO's planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) ensures acceptable mitigation of risk from any MRO-certified artificial intelligence modelsArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) and artificial intelligence applicationsArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) by considering all of the following: (1) The applicant's personnel and the qualifications of those personnel. (2) The quality of the applicant's planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) with respect to ensuring that artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) and application developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) exercise heightened care and comply with best practice-based standards for the prevention of personal injury and property damage, considering factors including, but not limited to, both of the following: (A) The viability and rigor of the applicant's evaluation methods, technologies, and administrative procedures. (B) The adequacy of the applicant's planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) to develop measurable standards for evaluating artificial intelligence developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c)' mitigation of risks. (3) The applicant's independence from the artificial intelligence industry. (4) Whether the applicant serves a particular existing or potential artificial intelligence industry segment.
(c) A designation as an MRO under this section shall expire after three years, and the MRO may apply for a new designation.
(d) The Attorney General may revoke a designation if any of the following are true: (1) The MRO's planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) is materially misleading or inaccurate. (2) The MRO systematically fails to adhere to its planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e). (3) A material change compromises the MRO's independence from the artificial intelligence industry. (4) Evolution of technology renders the MRO's methods obsolete for ensuring acceptable levels of risk of personal injury and property damage. (5) An artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) certified by the MRO causes a significant harm.
Section 8898.1 establishes the process by which the Attorney General designates private entities as MROs. The AG must evaluate whether the applicant's plan ensures acceptable mitigation of risk from MRO-certified AI models and applications, considering personnel qualifications, plan quality, evaluation rigor, independence from industry, and whether the applicant serves a particular AI market segment.
Designations expire after three years and are renewable. The AG may revoke a designation if the MRO's plan is materially misleading, the MRO systematically fails to adhere to it, the MRO's independence is compromised, technology evolves to render the MRO's methods obsolete, or a certified AI system causes significant harm. This section imposes obligations on the AG and on MRO applicants — not directly on AI developers.
(a)(1)–(10) 2 An applicant to the Attorney General for designation as an MRO shall submit with its application a planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) that contains all of the following elements: (1) The applicant's approach to auditing of artificial intelligence modelsArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) and artificial intelligence applicationsArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) to verify that an artificial intelligence developerDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) has exercised heightened care and adhered to predeployment and postdeployment best practices and procedures to prevent personal injury or property damage caused by the artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a). (2) The applicant's approach to mitigating specific high-impact risks, including cybersecurity, chemical, biological, radiological, and nuclear threats, malign persuasion, and artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) autonomy and exfiltration. (3) An approach to ensuring disclosure by developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) to the MRO of risks detected, incident reports, and risk mitigation efforts. (4) An approach to specifying the scope and duration of certification of an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a), including technical thresholds for updates requiring renewed certification. (5) An approach to data collection for public reporting from audited developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) and vendors that addresses all of the following: (A) Aggregating and tracking evaluation data from certified labs. (B) Categories of metadata to be aggregated and tracked. (C) Measures to protect trade secrets and mitigate antitrust risk from information sharing. (6) The applicant's intended use, if any, of security vendorsSecurity vendor"Security vendor" means a third-party entity engaged by an MRO or developer to evaluate the safety and security of an artificial intelligence model or application through processes that include red teaming, risk detection, and risk mitigation.Gov. Code § 8898(f) to evaluate artificial intelligence developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c), models, or applications, including a method of certifying and training vendors to accurately evaluate an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or developerDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) exercising heightened care and complying with best practices. (7) Implementation and enforcement of whistleblower protections among certified developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c). (8) Remediation of postcertification noncompliance. (9) An approach to reporting of societal risks and benefits identified through auditing. (10) An approach to interfacing effectively with federal and non-California state authorities.
(b) The planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) submitted pursuant to this section may be tailored to a particular artificial intelligence market segment.
(c) 3 An applicant shall annually audit all of the following to ensure independence from the artificial intelligence industry and report the findings of its audit to the Attorney General: (1) The applicant's board composition. (2) The availability of resources to implement the applicant's planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e). (3) The applicant's funding sources. (4) Representation of civil society representatives in evaluation and reporting functions.
(d) The Attorney General shall not modify a planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) submitted pursuant to this section.
Section 8898.2 prescribes the contents of the plan an MRO applicant must submit to the Attorney General. The plan must address ten specific domains: auditing approaches for pre- and post-deployment best practices; mitigation of high-impact risks including CBRN threats, malign persuasion, and AI model autonomy and exfiltration; developer disclosure of detected risks and incident reports; certification scope and duration including technical thresholds for recertification; data collection and public reporting; use of security vendors; whistleblower protections; remediation of noncompliance; societal risk and benefit reporting; and interfacing with federal and non-California authorities.
The section also requires annual independence audits covering board composition, resource availability, funding sources, and civil society representation. Notably, the AG is prohibited from modifying any submitted plan — ensuring the MRO retains full authority over its standards.
(a)–(c) 4 An MRO designated pursuant to this chapter shall do all of the following: (a) Certify developersDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c)' and security vendorsSecurity vendor"Security vendor" means a third-party entity engaged by an MRO or developer to evaluate the safety and security of an artificial intelligence model or application through processes that include red teaming, risk detection, and risk mitigation.Gov. Code § 8898(f)' exercise of heightened care and compliance with best practices for the prevention of personal injury and property damage. (b) Implement the planPlan"Plan" means a plan submitted pursuant to Section 8898.2.Gov. Code § 8898(e) submitted pursuant to Section 8898.2. (c) Decertify an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) that does not meet the requirements prescribed by the MRO.
(d) 5 Submit to the Legislature, pursuant to Section 9795, and to the Attorney General an annual report that addresses all of the following: (1) Aggregated information on capabilities of artificial intelligence modelsArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b), the observed societal risks and benefits associated with those capabilities, and the potential societal risks and benefits associated with those capabilities. (2) The adequacy of existing evaluation resources and mitigation measures to mitigate observed and potential risks. (3) DeveloperDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) and security vendorSecurity vendor"Security vendor" means a third-party entity engaged by an MRO or developer to evaluate the safety and security of an artificial intelligence model or application through processes that include red teaming, risk detection, and risk mitigation.Gov. Code § 8898(f) certifications. (4) Aggregated results of certification assessments. (5) Remedial measures prescribed by the MRO and whether the developerDeveloper"Developer" means a person who develops an artificial intelligence model or artificial intelligence application that is deployed in the state.Gov. Code § 8898(c) or security vendorSecurity vendor"Security vendor" means a third-party entity engaged by an MRO or developer to evaluate the safety and security of an artificial intelligence model or application through processes that include red teaming, risk detection, and risk mitigation.Gov. Code § 8898(f) complied with those measures. (6) Identified additional risks outside personal injury or property damage and the adequacy of existing mitigation measures to address those risks.
(e) 6 Retain for 10 years a document that is related to the MRO's activities under this chapter.
Section 8898.3 defines the ongoing operational responsibilities of designated MROs. MROs must certify developers and security vendors for compliance with best practices, implement their approved plans, decertify noncompliant AI models or applications, submit annual reports to the Legislature and Attorney General, and retain all documents related to MRO activities for ten years.
The annual report must cover aggregated AI model capabilities and associated societal risks and benefits, adequacy of existing evaluation and mitigation resources, certifications granted, aggregated assessment results, remedial measures and compliance therewith, and additional risks outside personal injury or property damage. This reporting obligation is significant — it creates a systematic public record of the AI safety landscape as observed through MRO auditing activities.
(a)–(b) 7 In a civil action asserting claims for personal injury or property damage caused by an artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a), it shall be an affirmative defense to liability that the artificial intelligence modelArtificial intelligence model"Artificial intelligence model" means an engineered or machine-based system that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.Gov. Code § 8898(b) or artificial intelligence applicationArtificial intelligence application"Artificial intelligence application" means a software program or system that uses artificial intelligence models to perform tasks that typically require human intelligence.Gov. Code § 8898(a) in question was certified by an MRO at the time of the plaintiff's injuries. (b) The affirmative defense recognized in this section shall not apply to claims of intentional misconduct by the defendant.
Section 8898.4 provides the bill's core incentive mechanism: in any civil action for personal injury or property damage caused by an AI model or application, MRO certification at the time of the plaintiff's injuries constitutes an affirmative defense to liability. The defense does not apply to claims of intentional misconduct by the defendant.
This safe-harbor provision is the principal reason developers would voluntarily seek MRO certification. It does not create an affirmative compliance obligation but modifies the litigation landscape for certified developers. The intentional-misconduct carve-out ensures the defense cannot shield deliberate wrongdoing.