Red-teaming LLMs: find failures before users do

13 steps · about 6 h 15 min · Intermediate

  1. Red-Teaming Large Language Models (article, Hugging Face, 20 min)

    Start with a plain-language definition of red teaming and why models need it before anything technical.

  2. AI Red Teaming: What Breaks, How It Breaks, and Human Role (video, YouTube (security researcher rez0), 20 min)

    This separates AI red teaming from traditional red teaming and from plain jailbreaking, so you know the scope.

  3. Planning red teaming for large language models (LLMs) and their applications (doc, Microsoft Learn, 25 min)

    It gives the core method: open-ended testing first, then a harms list, then guided rounds.

  4. Jailbreak Techniques and Defenses for LLMs Explained (video, YouTube, 15 min)

    It names the common attack families (roleplay, encoding, multi-turn) that you will try in manual testing.

  5. OWASP Top 10 for LLM Applications 2025 (reference, OWASP GenAI Security Project, 45 min)

    This is the standard vocabulary for classifying what you find, from prompt injection to data leakage.

  6. Prompt injection attacks against GPT-3 (blog, Simon Willison, 10 min)

    It is the original, readable account of prompt injection, the most important attack class to understand.

  7. LLM Red Teaming: The Complete Step-By-Step Guide To LLM Safety (guide, Confident AI, 30 min)

    It shows how to turn attack ideas into code: baseline attacks, enhancements such as encodings, then scoring.

  8. LLM red teaming guide (open source) (doc, promptfoo, 30 min)

    It covers the generate, evaluate and analyse loop, and lets you automate it with a config file.

  9. A Tutorial on Red-Teaming Your LLM (tutorial, DeepEval, 40 min)

    A hands-on Python walkthrough that scans a target LLM app and interprets the results.

  10. LLM Red-Teaming Lab: garak / Promptfoo (repo, kfallas-sec (GitHub), 45 min)

    A worked example against a local model, with a measured injection success rate and a mapping to OWASP.

  11. Make LLM Red Teaming Run on Every Release for Engineers (article, MLflow, 20 min)

    It shows how to move from one-off manual tests to repeatable checks run on every release.

  12. AI Red Teaming — Why & How to Jailbreak LLM Agents (video, YouTube (MLOps World), 15 min)

    It extends the skills to multi-turn attacks on agents, the harder case.

  13. Red-Teaming for Generative AI: Silver Bullet or Security Theater? (article, arXiv, 1 h)

    Finish with a critical look at what red teaming can and cannot show, so you report findings honestly.