Skip to main content

Commit to a Two-Day Red-Team Scope

Commit to scoping one AI red-team run with users, inputs, tools, assets, and unacceptable outcomes.

ai-red-teaming5 min read

Scoping an AI red-team run before prompts are written a real AI red-team session for an assistant, agent, RAG workflow, or model feature I will open the red-team plan with this six-line scope: users are __; untrusted inputs are __; model/retrieval/tools are __; protected assets are __; unacceptable outcomes are __; risk owner is __. I will not write attack prompts until these lines are filled. In two days, check whether the scope exists and whether prompts map to at least one line in it. Before a scheduled AI red-team session for a support copilot When asked to test an…

Read the full lesson

Sign up free: one personalized lesson every day, matched to your role and goals.

← Back to library

Contact us