ForesightSafety Bench is a comprehensive AI safety evaluation benchmark and leaderboard created by Beijing-AISI that organizes safety assessment into three hierarchical levels, 22 pillars and 94 fine-grained risk dimensions for frontier AI systems.
Endorsements made here support Beijing Institute of AI Safety and Governance (Beijing-AISI).
ForesightSafety Bench is a comprehensive AI safety evaluation benchmark and leaderboard created by Beijing-AISI that organizes safety assessment into three hierarchical levels, 22 pillars and 94 fine-grained risk dimensions for frontier AI systems.
Endorsements made here support Beijing Institute of AI Safety and Governance (Beijing-AISI).
Project Details
Updated 05/18/26 · By grantmaking.aiForesightSafety Bench provides a structured framework for evaluating AI systems across fundamental content safety, extended domains such as deception, embodied and environmental risks, AI-for-science safety, social impacts, and catastrophic and existential risk scenarios, as well as multiple industrial application areas. Beijing-AISI publishes open-source code, datasets and a public leaderboard for the benchmark, and a 2026 arXiv paper details how its three-level, 22-pillar, 94-dimension structure is used to diagnose safety weaknesses in state-of-the-art foundation models and inform governance and standards work.
People
Updated 05/18/26 · By grantmaking.aiPrincipal Investigator
Discussion
No comments yet. Be the first to share your thoughts.