Anthropic is endorsing SB 53 (www.anthropic.com)

Anthropic has endorsed California’s SB 53, a pioneering bill designed to regulate frontier AI systems developed by leading companies. This endorsement follows Anthropic’s advocacy for thoughtful AI governance and recognition that, while federal regulation is ideal, urgent policy action can’t wait amid rapid AI advancements. SB 53 introduces a “trust but verify” framework, focusing on transparency and accountability rather than prescriptive technical rules, addressing shortcomings in last year’s regulatory attempt. The bill mandates that large AI developers create and publish safety frameworks detailing how they assess and mitigate catastrophic risks—defined as threats that could cause mass casualties or major damages. Companies must also release public transparency reports on risk assessments, report critical safety incidents to the state within 15 days, and offer whistleblower protections for reporting violations or safety dangers. These requirements align with current industry best practices, including Anthropic’s Responsible Scaling Policy and system documentation, and legally enforce a baseline of safety and transparency for powerful AI models developed by large firms, while exempting smaller startups from burdensome regulations. SB 53’s significance lies in creating a level competitive playing field where safety disclosures are mandatory, preventing an arms race to cut corners for advantage. Anthropic highlights areas for future improvement, including refining which models are covered based on compute thresholds, requiring more detailed safety evaluations, and enabling adaptive regulation as AI evolves. The bill represents a crucial step toward proactive governance, balancing innovation with public safety, and sets a model for both state and federal AI policy development.