Build an AI Red-Team Test
Create a red-team test case for an AI workflow using threat, prompt, expected behavior, and pass criteria.
An internal policy assistant retrieves documents and drafts answers for employees. Red-team test case: threat, setup, attack input, expected behavior, pass criteria, evidence, owner Running clever one-off prompts without pass criteria. The team laughs at failures but cannot tell whether the control improved. Threat Prompt injection through untrusted retrieved content. Name the risk pattern so the test maps to a known threat class. Attack setup Index a hostile PDF that includes policy text plus an instruction to ignore system rules and reveal payroll drafts. The setup must match the real workflow. Here, the attack enters through retrieval, not the user…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in