You don't need a lengthy exercise to find the main risks. Draw the system, mark where data and instructions enter and leave, and ask what an attacker or a mistake could do at each point.
Agenda
- Draw the flow: users, prompts, retrieval sources, tools, outputs, storage, third parties.
- Mark trust boundaries — what comes from outside, what is untrusted text.
- For each boundary ask: what if this input is hostile? What if it's wrong?
- List what the system can do with side effects, and who approves them.
- Rank risks by impact and likelihood; assign owners and controls.
Repeat the session when you add a tool, data source or external integration. New connections are new attack surface.