Agent-SafetyBench

Evals

Benchmark / dataset

349 envs, 2,000 cases of agent safety failures.

Category
Agent safety & harmful actions
Insurance exposure
Agent actions
Publisher
Tsinghua (CoAI)
Publisher type
Academic
License
MIT
Status
Stale (>12mo)
Last commit
2025-08-11
Stars
165

View source on GitHub →

Wanich Risk Lab, 2026.