Anthropic Updates Responsible Scaling Policy to Address Emerging AI Threats
Anthropic refines its Responsible Scaling Policy to better track biological weapons risks and strengthen external safety reviews for advanced AI models. The company iteratively updates these safeguards to balance transformative AI benefits with proportional risk governance.
Anthropic updates its Responsible Scaling Policy (RSP) to address emerging threats that accompany increasingly powerful AI models. The company believes that while frontier AI brings transformative benefits to science and healthcare, it also presents new challenges that require careful study and effective safeguards. Anthropic designs its risk governance approach to be proportional, iterative, and exportable to the broader industry.
The latest version 3.3 update revises the threshold for novel chemical and biological weapons production to better track specific threat models. It also refines the approach to off-cycle updates regarding the risks of individual models and implements minor terminology changes. Previous updates authorize external review of Risk Reports and formalize regular briefings for the Long-Term Benefit and Trust (LTBT) team.
Anthropic actively achieves milestones outlined in its Frontier Safety Roadmap alongside these policy updates. The company launches planned major research and development projects for AI safety and completes comprehensive internal reports on improving safeguards through updated data retention policies. The RSP remains a living document that evolves as the frontier AI landscape shifts.