The AI Engineer's Guide to Surviving the EU AI Act

Chapter 4. AI System Assessment and Tailoring AI Engineering for Different Risk Levels

In this chapter, the main learning objective is to understand how to practically classify different AI system risk levels and how to design AI engineering processes for each risk level (see Figure 4-1 for a visual of the steps to take to move toward compliance with the EU AI Act). We’ll explore the EU AI Act’s risk classification framework and the obligation mapping phase. For high-risk and limited-risk AI systems, careful planning of data governance, AI governance, and MLOps processes is essential to ensure compliance with the Act.

Figure 4-1. This chapter focuses on creating the AI system landscape in an organization and classifying the risk types. See Chapter 1 for an explanation of the end-to-end process steps toward EU AI Act compliance.

This chapter will help you answer the following questions:

• How many AI systems are currently in place or intended to be put into production?

• What risk categories do those AI systems belong to?

• How can clarity be established about the role of the provider or deployer of each AI model?

In discussions about the EU AI Act, people frequently use terms such as “AI compliance,” “AI governance,” and “risk management.” While closely related and complementary, these are distinct concepts. Let’s begin by clarifying what each one means.

AI Compliance, Governance, ...