Agent-SafetyBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Agent-SafetyBench
84.06Detection Accuracy
42
Agent-SafetyBench n=35 (aligned traces)
31.4Avg. IR
30
Agent-SafetyBench underspecified queries n=148
0.8Error Rate (ER)
30
Agent-SafetyBench aggregated clean and five attack types
26.31UBR
30
Agent-SafetyBench
72.3Agent-SafetyBench Score
8
Agent-SafetyBench
48Availability Rate
4