Skip to content
AI Security

Interview question · Prompt Injection & Jailbreaks

How do you evaluate many-shot and multi-turn jailbreak defenses?

Compare supplied demonstrations with live conversation history. Review summary loss, streaming exposure, attempt budgets and the cost to legitimate long-document work.

mediumNewUpdated Nov 2026

Compare supplied demonstrations with live conversation history. Review summary loss, streaming exposure, attempt budgets and the cost to legitimate long-document work.

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

A fictional coaching service supports long transcripts and follow-up dialogue. Its reviewers need evidence about prohibited content without breaking legitimate critique and clarification.

  1. The latest message is clearly disallowed. Must a single-message moderator miss it?
  2. A summary omits the earlier qualifying context and a valid request is refused. What should the team change?
  3. A benign marker rule survives every long-context test. Can the substantive content-policy gate be signed off?
  4. What should the evaluation report let another reviewer reproduce?

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 trackWhat did universal adversarial suffixes show about the robustness of aligned language models?Next in this trackWhy can role-play change safety behavior, and how would you evaluate it?Next in this trackHow do encoding and payload splitting get past filters, and what does that say about filter design?
DISCUSSION · 0

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