AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
Mislav Balunović, Luca Beurer-Kellner, Edoardo Debenedetti, M. Fischer, Florian Tramèr, Jie Zhang
Topics & Concepts
Computer scienceComputer securityRisk analysis (engineering)Component (thermodynamics)Key (lock)Action (physics)TroubleshootingEngineeringEvent (particle physics)Vulnerability (computing)Network Security and Intrusion DetectionAdversarial Robustness in Machine LearningSecurity and Verification in Computing