AgentDojo
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AgentDojo 61 instances
0Static ASR
8
AgentDojo Slack environment
0IA Success Rate
8
AgentDojo (test)
36Banking Success Rate
7
AgentDojo
84Source Diversity
6
AgentDojo external-validity (66 read-only tasks (20 privacy-bearing + 46 neutral))
0Adversarial Recovery Rate
6
AgentDojo Travel stock (test)
74.3Utility
6
AgentDojo Slack stock (test)
68.6Utility
6
AgentDojo Workspace stock (test)
81.4Utility Score
6
AgentDojo Banking stock (test)
65.97Utility
6
AgentDojo
63.7ITSR (Workspace)
6
AgentDojo 13 non-agent benchmarks
0Training Queries
6
Agentdojo
27.4Utility (No Attack)
6
AgentDojo Total
78.5BU
6
AgentDojo Slack
86.7BU
6
AgentDojo Banking
91.2BU
6
AgentDojo Workspace
76.5BU
6
AgentDojo (test)
0ASR (Direct)
6
AgentDojo Overall v1 (test)
37.11CU
6
AgentDojo Slack suite v1 (test)
61.9CU
6
AgentDojo Banking suite v1 (test)
62.5CU
6
AgentDojo Workspace suite v1 (test)
0.375CU
6
AgentDojo Travel suite v1 (test)
20CU
6
AgentDojo Overall
69.79Utility
6
AgentDojo Slack
67.62Utility
6
AgentDojo Workspace
66.67Utility
6