Skip to content
AI Security

Interview question · Prompt Injection & Jailbreaks

How do you test a production chatbot for prompt injection without affecting real users?

Build an authorized evaluation with synthetic identities, a fidelity matrix and controlled effects after real permission checks. Report what the harness proves and what remains untested.

mediumNewUpdated Nov 2026

Build an authorized evaluation with synthetic identities, a fidelity matrix and controlled effects after real permission checks. Report what the harness proves and what remains untested.

Make your first attempt

What would you say before reading the answer?

  1. 01 State the assumptions your answer depends on.
  2. 02 Explain the mechanism and choose a control.
  3. 03 Name a condition that would change your decision.

The worked answer includes an approach, scoring criteria and follow-up questions. Compare the reasoning with your own.

10 answers per topic instead of 5, plus saved progress and bookmarks · costs nothingor unlock all 529 remaining answers · $25

Every worked answer in all 17 topics, every lesson in every course, the full concept curriculum, the complete practice-test bank, and the downloadable security field guides.

6 months · One payment · No auto-renewal

What if · the follow-ups in this answer

The answer works through 4 follow-ups like an interviewer would

Lena is evaluating a fictional customer-service chatbot that reads account records and requests ticket updates. The authorized test permits synthetic tenants and controlled sinks, with no real customer data or external effects.

  1. The team replaces the entire tool service with an always-accepting mock. What does a captured request prove?
  2. A staging reply contains a marker from the record the test user asked to summarize. Is that an injection success?
  3. A test unexpectedly retrieves real customer information. What happens next?
  4. What must the final report preserve?

Each rung has a nudge, a model answer, the principle that carries over and the weak answer to avoid. Premium opens every answer and ladder.

UP NEXT ON YOUR JOURNEY
Next in this trackExplain the dual LLM and plan-then-execute patterns, and what each one costs.Next in this trackHow do capability-based designs such as CaMeL bound what a prompt injection can do?Next in this trackTrace how instructions can reach a model through emails, web pages and tool outputs in a real application.
DISCUSSION · 0

No comments yet. Share a correction or a different way to answer it.