Anthropic Supports California's AI Regulation Bill SB 53

Anthropic announced its support for California's SB 53 bill, which regulates powerful AI systems. This decision builds on lessons learned from the previous regulatory attempt, SB 1047, and reflects the judgment that while federal-level regulation would be ideal, state-level action is necessary to keep pace with the speed of technological advancement.

SB 53 implements the principle of 'trust but verify' through disclosure requirements rather than technical mandates. The bill imposes the following obligations on large-scale AI developers:

• Develop and publish safety frameworks for managing catastrophic risks
• Issue transparency reports containing risk assessments and mitigation measures before model deployment
• Report major safety incidents to the state within 15 days
• Provide whistleblower protections for reporting violations and public safety risks
• Public accountability for compliance with frameworks, with financial penalties for non-compliance

These requirements align with Anthropic's existing practices, such as its Responsible Scaling Policy and the publication of system cards. Other major labs, including Google DeepMind, OpenAI, and Microsoft, have also adopted similar approaches.

SB 53 includes exemptions for startups and smaller companies to reduce unnecessary regulatory burden, while aiming to create a level playing field among large developers to prevent them from competing at the expense of safety.