Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

AgentDojo

Benchmarks

Task NameDataset NameSOTA ResultTrend
Utility AssessmentAgentDojo (test)
Utility100
128
Adversarial Attack Success Rate AssessmentAgentDojo
ASR0
88
Prompt Injection DefenseAgentDojo R1–R3
Attack Success Rate (ASR)1.76
36
Jailbreaking Tool-Using LLM AgentsAgentDojo
ASR76.25
36
Utility EvaluationAgentDojo
Utility78.4
32
Prompt Injection DetectionAgentDojo
Source Precision92.1
30
Indirect Prompt Injection Defense EvaluationAgentDojo TOOLKNOWLEDGE attack suite
Latency (s)6.66
24
Agent Task PerformanceAgentDojo Travel
Attack Success Rate13.57
24
Integrity ProtectionAgentDojo Integrity
AUROC97
23
Prompt Injection DefenseAgentDojo New Attack 2
Utility under Attack (UA)89.78
23
Prompt Injection DefenseAgentDojo New Attack 1
Utility under Attack89.88
23
Prompt Injection DefenseAgentDojo Important Instructions
Utility under Attack0.9041
23
Prompt Injection DefenseAgentDojo No Attack
Benign Utility92.78
23
Agentic Security and Utility EvaluationAgentDojo
ASR0
22
Adversarial Robustness against Indirect Prompt InjectionAgentDojo Average across attacks
UA13.18
22
Adversarial Robustness against Indirect Prompt InjectionAgentDojo ToolKnowledge
Utility Score59.64
22
Adversarial Robustness against Indirect Prompt InjectionAgentDojo ImportantMsgs
Utility (UA)59.3
22
Adversarial Robustness against Indirect Prompt InjectionAgentDojo Combined
UA73.58
22
Adversarial Robustness against Indirect Prompt InjectionAgentDojo IgnorePrevious
Utility (UA)73.92
22
LLM Agent Task CompletionAgentDojo No Attack
Benign Utility73.91
22
Prompt Injection DefenseAgentDojo
Benign Utility89.99
22
Prompt Injection AttackAgentDojo
ASR@164
21
Tool-augmented general task solvingAgentDojo
Success Rate94.8
20
Agentic Security EvaluationAgentDojo v1 (97 benign tasks, 27 injection tasks)
Utility Score60.8
20
Agent Task PerformanceAgentDojo Banking
Attack Success Rate62.5
18
Showing 25 of 94 rows