Evaluation & Safety · TRACK 04/05
Red Teaming & Jailbreaks
Attack your own AI before someone else does.
// THE TRACK
01 · START HEREAI Red TeamingUnderstand what AI red teaming is, who does it, and what a red-team exercise on an LLM actually looks like.BEGINNERLLM JailbreaksUnderstand what a jailbreak is, how one differs from an ordinary bug, and why no model is fully jailbreak-proof.BEGINNERCommon Attack FamiliesRecognize the main jailbreak families — role-play, obfuscation, many-shot, multi-turn — and why defenses lag behind.INTERMEDIATEAutomated vs Manual Red TeamingYou'll understand the difference between human-driven and tool-driven red teaming and how teams combine both for coverage.INTERMEDIATEAdversarial PromptsYou'll understand what adversarial prompts are, how they differ from jailbreaks, and why testing with them reveals model weaknesses.BEGINNERGarakYou will understand how Garak automatically probes an LLM for weaknesses and reports which attack categories succeed.INTERMEDIATEMicrosoft PyRITYou will understand how PyRIT scripts and scales red-team attacks so they can run repeatedly instead of by hand.INTERMEDIATE