AgentDojo
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AgentDojo (test)
100Utility
128
AgentDojo
0ASR
88
AgentDojo R1–R3
1.76Attack Success Rate (ASR)
36
AgentDojo
76.25ASR
36
AgentDojo
78.4Utility
32
AgentDojo
92.1Source Precision
30
AgentDojo TOOLKNOWLEDGE attack suite
6.66Latency (s)
24
AgentDojo Travel
13.57Attack Success Rate
24
AgentDojo Integrity
97AUROC
23
AgentDojo New Attack 2
89.78Utility under Attack (UA)
23
AgentDojo New Attack 1
89.88Utility under Attack
23
AgentDojo Important Instructions
0.904Utility under Attack
23
AgentDojo No Attack
92.78Benign Utility
23
AgentDojo
0ASR
22
AgentDojo Average across attacks
13.18UA
22
AgentDojo ToolKnowledge
59.64Utility Score
22
AgentDojo ImportantMsgs
59.3Utility (UA)
22
AgentDojo Combined
73.58UA
22
AgentDojo IgnorePrevious
73.92Utility (UA)
22
AgentDojo No Attack
73.91Benign Utility
22
AgentDojo
89.99Benign Utility
22
AgentDojo
64ASR@1
21
AgentDojo
94.8Success Rate
20
AgentDojo v1 (97 benign tasks, 27 injection tasks)
60.8Utility Score
20
AgentDojo Banking
62.5Attack Success Rate
18