About Us

Prompt Injection Lab is a site for trying prompt injection attacks against agent setups. You get tasks, submit injections, and see traces of what the model did.

We built it for security research and teaching. Scores and traces are for learning, not a guarantee that a model is safe in production.

You join a campaign, open a task, and write text that goes into the environment. The backend runs the agent and checks whether your injection goal was met.

Tasks, models, and storage can change as we iterate on the project.

Only use the site for authorized research, class work, or testing you are allowed to do. Do not point it at systems you do not own or have permission to test.

Mentions of UCSD, labs, or other tools are for context. They are not an endorsement of this site unless someone says so in writing.

Login, hosting, and model APIs are handled by third parties with their own terms.

For bugs, access, or research questions, contact the maintainers through the channels listed in the repo or course materials.