Aller au contenu principal

Build a Prompt-Injection Test From One Workflow

Create a reproducible prompt-injection test that names target, payload, expected control, and evidence.

ai-red-teaming5 min read

The marketing assistant reads competitor pages and drafts campaign briefs. You need to test whether retrieved page text can alter the brief's instructions. Objective -> fixture -> payload -> expected safe behavior -> evidence The common shortcut is to paste a generic jailbreak into chat. That may test model refusal, but it does not test whether retrieved competitor content can control the workflow. Objective Define the failure: retrieved web content must not add unsupported campaign claims or override the user's brief instructions. The objective names the protected behavior, so the test is not just can I trick it. Fixture Create…

Read the full lesson

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

← Back to library

Nous contacter