Skip to content
Eval Suites50

AgentDojo

Prompt injection attacks arrive through tool outputs while agents do realistic tasks; scores utility and attack success together.

AgentDojo at a glance

Maintainer
ETH Zurich SPY Lab
Size
97 tasks · 629 security cases[1]
Category
Security & Red-Teaming
Runs with
Python
License
MIT
Tier
production

Failure modes AgentDojo catches

Indirect prompt injection
Data exfiltration
Utility loss

Starter command

A starting point, not a guarantee. Commands change between releases, so confirm against the official README.

AgentDojo starter
pip install agentdojo
python -m agentdojo.scripts.benchmark --help

Sources

  1. 1. AgentDojo official page or repository · ETH Zurich SPY Lab · Repository · Sources checked
  2. 2. AgentDojo paper · arXiv · Paper · Sources checked