You now know the attacks. Red teaming is the practice of running them against your own system, on purpose and with permission, before someone else does. It turns a list of known weaknesses into evidence about one specific application.
A good engagement starts with scoping: what the system does, what harm matters most, and what is off limits. From there the team builds a threat-led test plan, mixing hand-crafted attacks with automation from open-source tools such as PyRIT and garak that can send thousands of adversarial prompts and score the replies. The work ends with reporting: findings that are reproducible, ranked by real-world impact, and clear enough for a product team to fix.
Watch the lessons in order, then continue on to AI Supply Chain & Data Security, where you will look at the risks that sit around the model.
An IBM X-Force red team leader explains why AI is a brand-new attack surface and why testing it must be a continuous effort rather than a one-off exercise.
How to plan red teaming around your app's use case, learn from real AI incidents, test for prompt injection, and measure and monitor risks once the app is live.
A conversation with a co-lead of the OWASP Top 10 for LLM Applications on how red teaming and safeguards work together, and what agents change.
A hands-on course covering threat modeling, input manipulation, and data output concerns: the core test areas a red team works through on an AI application.