On December 19, 2025, Anthropic published its Frontier Compliance Framework (FCF) to comply with California’s Transparency in Frontier AI Act (SB 53), which the company says will take effect on January 1. SB 53 establishes the United States’ first set of frontier AI safety and transparency requirements aimed at addressing catastrophic risks from the most capable AI systems.
What the FCF covers
According to Anthropic, the FCF explains how the company assesses and mitigates risks associated with its frontier models in several categories:
- cyber offense;
- chemical, biological, radiological, and nuclear (CBRN) threats;
- AI sabotage and loss of control.
The framework describes a tiered system for evaluating model capabilities against these risk categories and outlines the company’s mitigation approaches. It also addresses protections for model weights and procedures for responding to safety incidents.
Anthropic says much of the FCF reflects an evolution of practices the company has followed for years. Since 2023, Anthropic’s Responsible Scaling Policy (RSP) has articulated its approach to managing extreme risks from advanced AI systems and has guided development and deployment decisions. The company also publishes detailed system cards when launching new models that summarize capabilities, safety evaluations, and risk assessments.
Under SB 53, transparency practices of this kind become mandatory for organizations building the most powerful AI systems in California.
Relationship between the FCF and the RSP
Anthropic intends the FCF to serve as its compliance framework for SB 53 and other regulatory requirements. The Responsible Scaling Policy will remain the company’s voluntary safety policy, reflecting what Anthropic views as best practices even where those go beyond or differ from current regulations.
Anthropic’s call for a federal standard
Anthropic describes SB 53’s implementation as an important moment because it formalizes achievable transparency practices that responsible labs have been following voluntarily. The company argues that a federal AI transparency framework is now needed to ensure consistent standards across the United States.
Earlier in the year, Anthropic proposed a framework for federal legislation that emphasizes public visibility into safety practices while avoiding prescriptive technical mandates. Core elements of that proposal include:
- publishing a public secure development framework: covered developers should disclose how they assess and mitigate serious risks, including CBRN harms and harms from misaligned model autonomy;
- publishing system cards at deployment: documentation of testing, evaluation procedures, results, and mitigations should be publicly disclosed when models are deployed and updated if models are substantially modified;
- protecting whistleblowers: it should be explicitly unlawful for a lab to lie about compliance or to punish employees who raise concerns about violations;
- flexible transparency standards: requirements should be a minimum, adaptable set of standards that can evolve as consensus best practices emerge;
- limiting application to the largest model developers: to avoid burdening startups and smaller developers, requirements should apply only to established frontier developers building the most capable models.
Anthropic states that as AI systems grow more capable, the public deserves visibility into how those systems are developed and what safeguards are in place. The company said it looks forward to working with Congress and the administration to develop a national transparency framework that balances safety with continued U.S. leadership in AI.
Dates and next steps
- FCF publication date: December 19, 2025;
- SB 53 effective date (as cited by Anthropic): January 1 (the company’s statement accompanying the FCF);
- Responsible Scaling Policy in force since 2023 and retained as Anthropic’s voluntary safety policy.
The FCF is now available publicly as Anthropic’s compliance document for SB 53, while the company continues to press for flexible, federal-level transparency standards.
