Responsible AI Engineering: Alignment, Safety, and Governance

This course provides a practical, end-to-end exploration of responsible AI engineering for developers, researchers, and engineers working with modern AI systems. You’ll move from foundational concepts to applied techniques used to assess, align, and govern AI models in real-world deployments. The course begins by distinguishing AI safety from AI security and mapping the full spectrum of AI risks, including bias, robustness failures, misalignment, and misuse. You’ll then analyze why models fail by examining technical alignment breakdowns such as reward hacking and specification gaming. Through hands-on exercises, you’ll audit models using adversarial attacks and interpretability tools like LIME and SHAP, apply alignment methods inspired by RLHF and PPO-style optimization, and automate red-teaming workflows with PyRIT. The course concludes with advanced topics, including evaluating models for dangerous capabilities, implementing runtime governance, and constructing formal AI safety cases.