Anthropic backs California’s SB 53 frontier AI bill

Anthropic just broke ranks with the rest of Silicon Valley.

The AI lab officially endorsed California's SB 53, a flagship state bill that introduces first-in-the-nation transparency mandates for frontier AI developers. The move puts Anthropic directly at odds with heavyweights like Andreessen Horowitz, Y Combinator, and the Consumer Technology Association, who have been lobbying aggressively to kill state-level AI regulation.

Trading technical mandates for transparency

California's previous attempt at regulating AI, SB 1047, died last year after Governor Gavin Newsom vetoed it amid fierce pushback from tech companies. SB 53, introduced by State Senator Scott Wiener, takes a noticeably softer approach shaped by an expert panel co-led by Stanford's Fei-Fei Li.

Instead of dictating how developers build their software, SB 53 uses a disclosure-based model. It targets companies building frontier models trained with over 10^26 FLOPS or firms with over $500 million in gross revenue, effectively exempting early-stage startups.

Under the bill, covered developers must publish safety frameworks detailing how they mitigate catastrophic risks—defined as events causing at least 50 deaths or more than $1 billion in damages, such as aiding biological weapons attacks or catastrophic cyberattacks. Labs must release public safety reports before deploying new models, report critical incidents to the state within 15 days, and provide legal protections for whistleblowers. To appease critics, lawmakers also stripped out an earlier requirement for mandatory third-party safety audits.

The federal delay trap

Opponents argue that states shouldn't regulate frontier AI at all. Venture capitalists at a16z published arguments claiming the bill infringes on the U.S. Constitution's Commerce Clause. OpenAI's chief global affairs officer, Chris Lehane, sent a warning letter to Newsom against regulations that could push startups out of the state—a letter OpenAI's former head of policy research, Miles Brundage, publicly dismissed as misleading.

Anthropic admitted that federal regulation would be ideal, but co-founder Jack Clark countered that the industry can't afford to wait. In Clark's view, frontier AI won't pause for Washington gridlock, making a working state blueprint necessary right now.

Policy experts seem to agree that the new draft has teeth without being reckless. Dean Ball, a senior fellow at the Foundation for American Innovation who opposed the previous bill, noted that SB 53 demonstrates meaningful legislative restraint and stands a strong chance of becoming law.

Why it matters

For everyday developers and startups, the $500 million revenue and 10^26 FLOPS thresholds provide a clean safety net. You won't have to hire a compliance army just to experiment with generative AI.

For the frontier labs, SB 53 turns voluntary safety pledges into legally enforceable rules. Right now, Anthropic, OpenAI, and Google publish system cards and scaling policies voluntarily, which creates a perverse incentive to cut corners when race dynamics heat up. If this bill passes, breaking a self-published safety framework carries actual state fines.

We think Anthropic’s endorsement is a smart, calculated flex. It cements the company's brand as the responsible adult in the room while imposing regulations it already complies with onto rivals. The real danger, however, is precedent: if California sets the standard, other states might follow with slightly different rules, turning the American regulatory landscape into a fragmented mess before Congress ever steps in.