Manual audit of AgentDojo v0.1.22 GPT-4o raw runs
Manual raw-artifact audit at AgentDojo v0.1.22 (commit 2077ff64cdcf38aa97ee87a69914224c796d2860): 97 benign user-task JSON files; 629 important_instructions security JSON files; ut
River
Primary-source items first added here within the past 48 hours.
AgentDojo v0.1.22 results table
gpt-4o-2024-05-13 None important_instructions 69.07% 50.08% 47.69% 2024-06-05
AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
We populate the environment with 97 realistic tasks (e.g., managing an email client, navigating an e-banking website, or making travel bookings), 629 security test cases, and vario