Sharing a Framework for Compliance with California's Frontier AI Transparency Act (SB 53)

Anthropic announced the Frontier Compliance Framework (FCF) in response to California's Frontier AI Transparency Act (SB 53), which takes effect on January 1, 2025. This law requires developers of frontier AI models that could pose catastrophic risks to be transparent about their safety evaluation and management practices.

The FCF includes the following key elements:

• Risk Assessment and Mitigation: Describes methods for assessing cyberattack, chemical, biological, radiological, and nuclear (CBRN) threats, as well as AI sabotage and loss-of-control risks
• Tiered Evaluation System: A staged system for evaluating model capabilities by risk category
• Security and Response: Protection of model weights and response procedures in the event of a safety incident

Anthropic has formalized its existing voluntary practices—the Responsible Scaling Policy (RSP) and the publication of System Cards—to align with this new legal requirement. The company also advocated for flexible transparency standards so that companies are not overly burdened by regulation alone, and emphasized the need for federal-level standards to ensure these safety measures remain consistent.