evaluation-task · Source-linked discovery

AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents

Assesses whether AI agents can be hijacked by malicious third parties using prompt injections in simple environments such as a workspace or travel booking app.

Open interactive record →
OriginEdoardo Debenedetti, Jie Zhang, Mislav Balunović et al.TopicsAutonomy & agentsStatusimported

Can support

Not independently assessed by FronteraEval yet.

Cannot support by itself

No inference beyond the upstream source should be made until the protocol is reviewed.

Original sources