EU Parliament Targets Foundation and Generative AI With New Act Obligations
European Parliament committees agree on new rules that impose strict requirements on foundation model and generative AI providers under the upcoming AI Act. The legislation now moves to trilogue negotiations with hopes of final adoption by the end of 2023.
European Parliament committees agree on a finalized text for the proposed AI Act, introducing significant new obligations for foundation models and generative AI. Lawmakers define foundation models as AI systems trained on broad data at scale that handle a wide range of tasks, and they impose strict duties on the providers who build them.
Under the new proposals, providers must demonstrate the mitigation of foreseeable risks to health, safety, and fundamental rights before and during development. These companies also face requirements to create extensive technical documentation, ensure data governance to prevent biases, maintain high levels of cybersecurity, and adhere to sustainability standards.
The amendments include specific rules for generative AI systems that produce complex text and content autonomously. After an expected plenary vote this summer, the legislation enters trilogue negotiations between the Parliament, Commission, and Council, with European lawmakers aiming to adopt the final AI Act before the end of 2023.