A version of the following comments was submitted to the California Assembly Committee on Privacy and Consumer Protection on June 30, 2026.
Senate Bill 813 (SB 813) would establish the California Artificial Intelligence Standards and Safety Commission. This commission would develop a voluntary state framework to guide the implementation of voluntary safety standards for AI systems while also supporting voluntary audits and pilot programs to provide policymakers and practitioners with better information about AI risks and performance. The bill would not itself impose new mandatory safety obligations on all AI developers, would not create automatic liability, and would not grant the commission direct enforcement powers. Instead, it provides tools and reference points that other laws, agencies, and procurement policies may choose to use to make AI oversight more consistent and informed.
While SB 813 contains several constructive ideas, the types of standards and auditing framework it envisions are better suited to the federal level than to California alone. AI systems are designed, trained, and deployed across state lines, and the same model can serve users in every state at once, so rules that shape AI governance are inherently interstate rather than purely local. A national approach can provide a single, coherent set of expectations rather than having multiple state commissions issue overlapping and possibly conflicting guidance.
The bill points in a useful direction, but many of those same proposed tools can and should be organized around existing federal work, especially the AI Risk Management Framework and related guidance deployed by the National Institute of Standards and Technology (NIST), as well as recommendations from the Center for AI Standards and Innovation (CAISI). Aligning with those efforts would give companies a single reference point for risk management practices and evaluations rather than forcing them to navigate different state models layered on top of federal guidance. That sort of alignment requires congressional action to establish a national framework that states can plug into, rather than 50 potentially juxtaposed systems.
A federal framework could incorporate concepts similar to minimum and advanced safety standards to provide clearer expectations and support a shared audit and evaluation ecosystem that uses common terminology, metrics, and documentation practices. That would make it easier for firms to design internal governance once and then apply it consistently across the country, rather than customizing it state by state.
Advanced safety standards would provide firms with a clear benchmark for going beyond the minimum or competing on safety and could encourage stronger testing, monitoring, and risk mitigation by giving companies a more credible way to demonstrate those efforts to regulators, customers, and partners. These benefits depend on keeping standards and accreditation voluntary, so they can improve information and create reputational incentives without turning immediately into rigid legal mandates. A national approach will also better protect smaller firms, startups, and open source projects from cumulative compliance burdens. Adapting to one federal framework is demanding but manageable. Adapting to a growing set of state commissions with their own expectations would be much harder and risks entrenching larger incumbents that can absorb the cost of tracking and meeting many different standards.
SB 813 highlights useful tools for AI governance, but they would be more effective as part of a single national roadmap rather than many separate state maps. A federal system, grounded in existing national guidance and established by Congress, can offer a coherent framework for standards, evaluations, and voluntary audits so companies are not constantly replotting their course as they cross state lines. That kind of unified roadmap would provide a clearer direction, reduce the risk of contradictory routes, and better reflect the interstate nature of AI, while keeping standards and accreditation mechanisms voluntary rather than immediately coercive.