How to Red Team GenAI: Challenges, Best Practices, and Learnings

**Executive summary:** Generative AI red-teaming guidance from Microsoft outlines how GenAI testing differs from traditional red teaming due to probabilistic outputs, multimodal architectures, and responsible-AI risks; it recommends combining manual red teaming with automation using the open PyRIT framework to generate harmful prompts, adapt tactics, and identify risk hotspots, while emphasizing collaboration and best practices to mitigate data leakage, bias, hallucinations, and misuse.