Anthropic updates Transparency Hub detailing voluntary AI safety commitments
Anthropic refreshed its Transparency Hub page outlining progress on voluntary commitments, including its Responsible Scaling Policy and new Frontier Compliance Framework for regulatory requirements like SB 53.
Summary
Anthropic's Transparency Hub page, last updated on 23 July 2026, sets out how the company says it is meeting and working toward its voluntary AI safety commitments. The company says its experience with several voluntary frameworks has shown consistent themes and considerable overlap on safety, security and responsible development, and it organises the overview by focus area while inviting feedback from the AI community and policymakers.
On risk, the page points to the Responsible Scaling Policy, first published in September 2023 and updated in February 2026. It ties safeguards to risk, with stronger protections as capability thresholds are reached. Its thresholds address chemical and biological weapons, misaligned systems in high-stakes uses, and automated R&D. A Frontier Compliance Framework, published in December 2025, addresses cyber offense, CBRN threats, sabotage, loss of control and harmful manipulation, and serves as the company's compliance framework for SB 53. Assessment methods include regular evaluations, threat modelling, red teaming, expert consultation, pre-deployment testing and outside evaluations by the UK AI Security Institute, US CAISI and METR. The company also reports ISO/IEC 42001:2023 certification for its AI management system.
The page lists post-deployment measures such as a responsible disclosure policy, bug bounty programmes, user safety reporting and whistleblowing channels for employees. It describes cybersecurity safeguards, including a Cyber Verification Program, Claude Security and supply chain requirements, and notes that customers keep rights to their inputs and outputs and may opt out of training. Transparency measures include system cards and Risk Reports published every three to six months. The page also describes Anthropic's public benefit corporation status and its Long-Term Benefit Trust, its role in the Frontier Model Forum, and a four-year partnership with the Gates Foundation worth $200 million.