Update to Anthropic’s Responsible Scaling Policy: Enhancing Risk Governance for Frontier AI Systems

Anthropic Releases Significant Update to Responsible Scaling Policy for AI Risk Governance

Anthropic, a leading AI research company, has announced a major update to its Responsible Scaling Policy (RSP), a risk governance framework designed to mitigate potential catastrophic risks from frontier AI systems. The updated policy introduces a more flexible and nuanced approach to assessing and managing AI risks while maintaining a commitment to implementing adequate safeguards before training or deploying models.

The advancements in the RSP include new capability thresholds to indicate when upgraded safeguards are necessary, refined processes for evaluating model capabilities and safeguard adequacy, and new measures for internal governance and external input. Drawing inspiration from safety case methodologies and risk management practices in high-consequence industries, Anthropic aims to better prepare for the rapid advancement of AI technology.

Frontier AI models have the potential to bring transformative benefits to society and the economy, such as accelerating scientific discoveries, revolutionizing healthcare, and creating new domains for human creativity. However, these systems also present new challenges and risks that require careful study and effective safeguards.

The updated policy focuses on catastrophic risks like autonomous AI research and development and the potential misuse of AI in creating chemical, biological, radiological, and nuclear weapons. Anthropic has defined specific Capability Thresholds that would trigger the need for upgraded safeguards, such as elevated security standards and deployment controls.

To ensure effective implementation of the policy, Anthropic has established capability assessments, safeguard assessments, documentation and decision-making processes, and measures for internal governance and external input. The company has also learned valuable lessons from its first year of implementing the RSP, leading to improvements in flexibility and compliance tracking.

As the frontier of AI continues to advance rapidly, Anthropic remains committed to evolving its safety program, policies, evaluation methodology, safeguards, and research into potential risks and mitigations. Co-Founder and Chief Science Officer Jared Kaplan will now serve as Anthropic’s Responsible Scaling Officer, overseeing the continued implementation of the RSP.

Anthropic is also hiring for a Head of Responsible Scaling to coordinate efforts in complying with the policy. The company invites contributions to AI risk management and encourages interested individuals to apply for various roles within their risk management teams.

For more information on the updated Responsible Scaling Policy, visit anthropic.com/rsp and supplementary information at anthropic.com/rsp-updates. Anthropic extends gratitude to external groups for their feedback on the development and refinement of the policy.