The EU regulation on artificial intelligence is entering its final phase with the adoption by the EU Parliament of its position on the AI Act on 14 June 2023.
We have discussed the risk-based approach of such draft regulation in a previous article (here). Another important feature concerns specific AI systems which are to receive specific rules depending on their characteristics.
The EU Parliament proposes to adopt a technology-neutral approach and differentiate (i) general-purposeAI systems and (ii) foundation models to impose strict requirements for each category.
General purpose AI systems are a type of AI that can be used for applications in many disciplines such as finance or healthcare. General purpose AI systems can be used to leverage other AI systems that were not designed for such purpose. This may result in breaches of existing regulations such as the EU GDPR.
Foundation models can be used for a large range of different tasks (e.g. GPT-3). The draft regulation will impose an obligation on providers of foundation modelsto guarantee an effective protection of data subjects and users, health, safety and democracy. This will require specific monitoring from foundation models providers.
Regarding generative AI models using large language models or broad datasets to create all types of contents, such as Chat GPT or MidJourney, specific transparency requirements will apply.
The draft AI act is currently designed to have an extra territorial effect regarding third countries to the EU such as Monaco. Overseas AI system providers are recommended to carefully monitor the finalization of this regulation.
DL Corporate & Regulatory regularly assists tech companies with respect to the rolling out of innovative and AI-based projects. Vincent de Bonnafos, one of our lawyers is a member of DL4T a research project in Nice University leading in the field of regulations applicable to AI and deep technologies.