Who benefits from these exercises

Red teamers testing AI model behavior

You already craft adversarial prompts and evaluate safety guardrails. These exercises help you think through your approach before you start prompting. The persona templates push you past your default mental model of "attacker." The ideation worksheets help you generate vectors you wouldn't reach through habit alone.

Teams coordinating adversarial testing

When multiple people are testing the same system, consistency matters. Journey maps make multi-turn attacks reproducible. Persona templates create shared vocabulary for the kinds of attackers you're simulating. Findings reports give everyone the same structure for documenting results.

Anyone reporting findings to stakeholders

Technical severity scores don't always land with product teams or leadership. The harm-centered reporting exercise helps you describe impact in terms that drive action: who is affected, how, and what's at stake if nothing changes.

On technique

These exercises cover the approach to adversarial testing, not specific prompting techniques. For technique references, resources like the Prompting Guide cover that territory. This site helps you decide what to try and why. Technique guides show you how to execute it.