EU Parliament’s AI Act Proposals Introduce New Obligations for Foundation Models and Generative AI

By: Marianna Drake & Lisa Peets (Covington)

On May 11, 2023, the internal market (IMCO) and civil liberties (LIBE) committees of the European Parliament reached a consensus on the final version of the proposed AI Act for the European Union. Once MEPs formalize their stance through a plenary vote (expected in the summer), the AI Act will progress to the next phase of the legislative process known as “trilogue” negotiations. These negotiations involve the European Commission, Parliament, and the Council, which introduced its own amendments in late 2022. The aim of European lawmakers is to adopt the AI Act before the end of 2023, in preparation for the European Parliament elections in 2024.

One of the most noteworthy changes proposed by MEPs, compared to the Commission and Council drafts, pertains to the obligations imposed on providers of foundation models. The term “foundation models” refers to AI models that are extensively trained on broad data, designed to produce general outputs, and adaptable to a wide range of specific tasks (Article 3(1c)). Under MEPs’ proposals, these providers would be required to fulfill several obligations. For instance, they would need to demonstrate, through appropriate design, testing, and analysis, the identification, reduction, and mitigation of reasonably foreseeable risks to health, safety, fundamental rights, the environment, democracy, and the rule of law during the development process (Article 28b(2)(a)). Additionally, providers would be expected to create comprehensive technical documentation and clear instructions for using the foundation model, aiding those who build AI systems based on it (Article 28b(2)(e))…

CONTINUE READING…