An overview of California's AI safety bill SB 1047, which easily passed the state's legislature, as it awaits Governor Newsom's signature or veto before Sep. 30
California's Gavin Newsom has until September 30 to decide whether to sign legislation that will reach far beyond the state
Context & Ripple Effects
SB 1047 reached the governor after the State Assembly advanced it, completing the legislature’s role in a closely watched state-level attempt to set AI-safety obligations. The immediate issue was whether Newsom would turn that legislative consensus into a statewide standard.
The later arc underscores the significance of the decision point: Newsom vetoed SB 1047 over its model-focused scope, while California subsequently enacted SB 53’s safety-testing disclosure regime. That sequence points to a shift in regulatory design rather than an end to California AI oversight.
First-order effects
- Newsom’s signature or veto determines whether SB 1047 becomes an operative compliance obligation; until then, affected AI companies face policy uncertainty rather than a new enforceable state requirement.
- The bill’s passage forces AI developers, safety advocates and state policymakers to focus on its scope and implementation trade-offs before the deadline.
Second-order effects
- A veto would push advocates toward a narrower or differently structured proposal, while a signature would make California a practical reference point for companies serving the state and for other state legislatures.
- The debate sharpens the distinction between rules keyed to large models and rules keyed to risky deployment—a concern later central to Newsom’s stated veto rationale.
Third-order effects
- California’s subsequent move to require disclosure of safety-testing regimes suggests durable AI governance may develop through auditable process requirements rather than a single model-size threshold.
- If states continue to adopt different AI rules, developers may increasingly build governance and documentation systems that can satisfy the most consequential state regimes, though the eventual degree of convergence remains uncertain.
The trend: This is one step in the maturation of state-level AI governance from broad safety mandates toward operational, disclosure-based oversight.