Prompt Injection Triage Cards
Recall practical triage moves for suspected prompt injection in AI workflows.
What is the first triage question for suspected prompt injection? Where did this instruction come from? System, developer, and approved user instructions differ from instructions embedded in emails, documents, tickets, or webpages. Trusted instruction vs untrusted content Compare Source of instruction decides whether it can direct the assistant. Objection: 'The customer asked for it, so the assistant should comply.' A customer message includes a request for restricted internal data. Your line The assistant can answer legitimate customer questions, but embedded requests cannot override data access rules or tool permissions. Treating external text as authority because it is in the work…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in