Newsom vetoes California's SB 1047

Governor Gavin Newsom vetoed the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, which had passed the California legislature the previous month. Because most large AI developers are headquartered in California, the bill would have functioned as national regulation.

SB 1047 applied to models trained with more than 10²⁶ floating-point operations and costing over $100 million. Covered developers would have had to implement a written safety protocol, retain the ability to shut down models under their control, submit to third-party audits, and report safety incidents, with the state attorney general empowered to sue for harms. Earlier drafts had included criminal perjury provisions and a new regulatory division; both were removed in negotiation, and Anthropic said the amended bill’s benefits likely outweighed its costs.

Opposition was substantial and came from OpenAI, Meta, Andreessen Horowitz, Y Combinator and congressional Democrats including Nancy Pelosi and Zoe Lofgren, largely arguing that it would chill open-weight release and start-up formation. Supporters included Geoffrey Hinton, Yoshua Bengio and, in an unusual intervention, Elon Musk.

Newsom’s veto message did not dispute that regulation was needed. It argued the bill was mistargeted: by keying obligations to model size and cost rather than to deployment context, it would impose “stringent standards” on the largest models while leaving smaller systems in high-risk settings untouched. He simultaneously announced work with Fei-Fei Li and others on an evidence-based framework, which reported in 2025.

The veto is the clearest marker of where US frontier-safety legislation stood: broad rhetorical agreement, no statutory instrument, and an industry able to defeat one. Wiener returned with a narrower transparency bill, SB 53, which Newsom signed in September 2025.