RAI

With an exponentially increasing number of AI applications, RAI is becoming more important than ever before. While there are multiple definitions for RAI, they’re all related to the principles of fairness, interpretability, privacy, and safety in AI-based applications. [14]

Given the vast capabilities of generative AI models and their random, non-deterministic nature, establishing these principles challenges architects. As this is an emerging discipline, there are no established frameworks such as ISO 27001 or NIST that can guide practitioners. For this purpose of this book, we’ll loosely follow the recently published EU AI Act, which will become mandatory for (generative) AI system providers in August 2026 (yes, that probably includes you!). Please keep in mind that this book doesn’t provide legal guidance and that you should carefully review any system design with legal experts and your security and governance teams before you make it available to the...