AI models can do remarkable things. They can also behave in ways their creators never intended.

This course takes you inside some of the most important failure modes in modern AI. You’ll explore guardrails, jailbreaks, prompt injection, leaking, poisoning, toxicity, and hallucinations while seeing how these issues show up in real LLM interactions.

You’ll go beyond definitions, too. Try creating jailbreak prompts, examine multimodal injection, investigate hallucinations through mechanistic interpretability, and compare how different models think and fail.

Finally, you’ll learn how model cards document capabilities, limitations, and risks—and practice reading them with a more critical eye.

By the end, you’ll have a clearer mental model of where AI systems can break, why those failures matter, and how AI safety researchers evaluate them.