OpenAgentSafety

Evals

Benchmark / dataset

Real-tool agent safety tasks (file, browser, code, messaging).

Category
Agent safety & harmful actions
Insurance exposure
Agent actions
Publisher
CMU
Publisher type
Academic
License
MIT
Status
Active (<90d)
Last commit
2026-08-10
Stars
37
Notes
ICLR 2026; moved from sani903/OpenAgentSafety Aug 2026.

View source on GitHub →


Wanich Risk Lab, 2026.