Secure by Design and Trustworthy AI

We have covered key examples of adversarial AI attacks in significant detail. This chapter transitions from understanding the adversarial landscape of AI to embedding security and trustworthiness into AI systems from the design phase.

While we’ll focus on the technical details of attacks to help us understand them, we will take a step back and try organizing our examples and understanding around more formal definitions of threats and attacks by using categories and taxonomies devised by leading organizations such as NIST, MITRE, and OWASP. This will provide a foundation to reference the threats and attacks without getting lost in the details new variants will no doubt bring.

We will consolidate these taxonomies into a reference model that we can use to plan and deliver Trustworthy AI, a term that incorporates security and safety. We will discuss the vital role of data and AI ethics in Trustworthy AI and how they relate to security and...