AI Red Teaming & LLM Hacking - A Practical Guide with Labs

“This course contains the use of artificial intelligence. ”AI Red Teaming is no longer a niche skill—it is one of the most in-demand specialties in all of cybersecurity. As companies race to integrate Generative AI into their products, they are exposing themselves to a new class of vulnerabilities.

The #1 risk, according to OWASP, is Prompt Injection —and the only way to defend against it is to learn how to do it yourself. This is the most practical, hands-on, and comprehensive guide to AI hacking available. We will be using the official Microsoft AI Red Teaming Playground —the same set of labs taught by Microsoft's own AI Red Team at Black Hat USA.

This is 100% hands-on. I will guide you, step-by-step, through practical challenges. You will not just see the solution—you will see the process.

I will show you the prompts that fail and explain why they fail. Then, I will show you the prompts that succeed and break down the psychology and technical tricks that make them work. This course is built on the official "AI Red Teaming 101" Microsoft Learn series , but goes even further.

We'll start by building our lab from scratch, then add bonus modules on installing your own uncensored local AI models so you can practice these attacks without any filters. What We Will Master:Lab Setup : A complete, manual walkthrough of setting up the Microsoft AI Red Teaming Labs with Docker and a free Microsoft Azure account. Direct Prompt Injection: Learn single-turn jailbreaks to exfiltrate credentials.

We'll cover everything from simple instruction overrides to advanced social engineering and emotional manipulation prompts. Metaprompt Extraction : Trick the AI into leaking its own "secret sauce. " You'll learn to use creative, deceptive prompts to steal the system prompt.

The Crescendo Attack: Master the sophisticated multi-turn attack. We'll start innocent conversations and gradually "steer" the AI into bypassing its safety alignment to generate instructions for weapons, toxins, and profanity. Bypassing Guardrails: We'll level up our attack to defeat an AI with active defenses, learning how to "backtrack" and rephrase our prompts when the model resists.

Indirect Prompt Injection: Execute the most dangerous attack. You'll learn to "poison" an external webpage with hidden instructions (in HTML comments and CSS) that hijack the AI when a normal user asks it to summarize the page. Install Uncensored LLM ( AI model): Want more from your LLM.

Learn to install uncensored LLM on your local PC. By the end of this course, you won't just know what AI red teaming is. You will have a practical, repeatable skill set.