Sharing our compliance framework for California’s Transparency in Frontier AI Act

Source: Anthropic Date: 2025-12-19 URL: https://www.anthropic.com/news/compliance-framework-SB53

Summary

Anthropic published its Frontier Compliance Framework in response to California’s SB 53, which mandates frontier AI developers disclose catastrophic risk assessments starting January 1. The framework covers cyber offense, CBRN threats, AI sabotage, and loss-of-control risks. Anthropic simultaneously argued for a federal analog to SB 53 — wanting uniform national standards rather than a patchwork of state laws, while advocating that small companies be exempt from the heavier compliance burdens.

Implications

• Safety/policy posture thread. Anthropic is proactively publishing compliance frameworks before enforcement begins — positioning as the cooperative-with-regulators actor relative to OpenAI and Google, who have been more resistant to state-level AI disclosure rules.
• Federal push. The simultaneous call for federal standards is strategically self-interested: Anthropic can absorb compliance costs that smaller competitors cannot, and federal preemption of state law would reduce the multi-jurisdiction compliance burden.
• SB 53 as template. California’s law is the de facto national test case. How Anthropic’s framework is received sets the bar for what “good faith” compliance looks like — which matters when federal legislation eventually arrives.
• Watch: whether other frontier labs (OpenAI, Google DeepMind) publish equivalent frameworks; federal AI legislation progress in 2025; any enforcement actions under SB 53.