← all datasets

AgentHazard

Emerging
1papers using it
2026first seen

AgentHazard is a benchmark that evaluates the safety of computer-use agents by assessing their performance on trajectory-level tasks that simulate multi-step execution traces, where individual actions may seem benign but can lead to harmful outcomes.

Papers using AgentHazard (1)

AgentHazard β€” datasets β€” cybersecurity