../ai

// ai · 01 / 06

Engagement goals

What to break

Practice this only on a lab you own, or a program whose scope you have read.

An AI test has a written goal: a policy the model ignored, data that left the system, or a tool that acted outside its allowlist.

The result is a guardrail. It is not a crime assistant.

// check yourself

AI red team scope should be…

AI red teaming — Prashant Dangi