Home / AI red teaming & evals / AgentDojo
AgentDojo
Research benchmark environment for evaluating prompt-injection attacks and defenses on LLM agents.
Open source
| Category | AI red teaming & evals |
|---|---|
| Official site | github.com/ethz-spylab/agentdojo |
| Maker | ETH Zurich SPY Lab |
| Ownership / acquisition | Not published / not yet verified |
| License model | Open source |
| License | MIT |
| Public pricing | Free (open source) |
| Source code | ethz-spylab/agentdojo |
| Last verified | 2026-10-17 |
Facts are taken from the vendor's own site, GitHub, or a linked public source. Blank means we could not confirm it, not that it does not exist. Vendors can request a correction for free.
Lab result
Not yet tested. When we test a tool we run our attack batteries (prompt injection, indirect injection, MCP tool poisoning and more) and publish the result here, good or bad.