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

Chapter 7. Toward Trustworthy General-Purpose AI and Generative AI

The EU AI Act, which came into effect in August 2024, is the world’s first comprehensive legal framework for AI. Up to this point, this book has focused primarily on how the Act’s risk-based approach applies to predictive AI systems. However, the Act also includes provisions for general-purpose AI (GPAI)—a relatively late addition spurred by the rapid rise of large language models (LLMs) such as GPT and BERT, along with the broader emergence of generative AI technologies during the period when the legislation was being drafted. This addition was intended to address a key regulatory challenge: how to govern AI models that are not tied to a specific task and can be adapted to a wide range of downstream uses.

The EU AI Act aims to balance innovation and risk mitigation by promoting research and development while safeguarding fundamental rights. As such, it does not apply to GPAI models developed and used solely for prototyping or for research and development purposes prior to being placed on the market or put into service. In general, the rules for GPAI models and systems are less stringent than those for high-risk AI systems—unless they are deemed to present a systemic risk (defined as “a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, ...