AI Red Teaming Foundations with PyRIT and HarmBench

As language models are integrated into critical applications, ensuring their safety and robustness against adversarial inputs is more important than ever. This course introduces you to the fundamentals of automated AI red teaming, helping you identify vulnerabilities and secure your models before they reach production.\n\nYou will learn how to set up automated safety evaluation pipelines to systematically test model boundaries and calculate key safety metrics.\n\nWhat you'll learn:\n- Understand foundational AI safety concepts, vulnerability types, and the role of red teaming\n- Configure PyRIT to automate the generation of adversarial test cases for language models\n- Apply HarmBench to systematically evaluate model robustness against standardized benchmarks\n- Calculate Attack Success Rate to measure and track model safety over time\n- Design basic mitigation strategies, including system prompt engineering and guardrails, to address identified risks\n\nYou will start with core terminology and safety definitions before moving on to the setup and configuration of automated evaluation frameworks. The course concludes with practical guidance on interpreting safety metrics and implementing protective guardrails.\n\nThis course is designed for software engineers, data scientists, and security beginners interested in AI safety, requiring no prior experience with red teaming tools.\n\nStart building safer AI systems today.