EU AI Act Greenlit: Comprehensive Guide to the Artificial Intelligence Regulations in Europe

The European Union achieved a significant milestone by endorsing the AI Act, marking the first legislative effort to govern the widely debated realm of artificial intelligence (AI), which holds the potential to transform our daily lives. Following an extensive three-day negotiation involving the trialogue with the European Commission, Council, and Parliament, the AI Act was sanctioned on December 8, 2023, to oversee the use of AI systems within the European Union.

The approval of the AI Act was not guaranteed until recently. The French, German, and Italian governments had advocated replacing it with a mere code of conduct to reduce regulatory burdens on European companies, aiming to enhance their global competitiveness. However, European legislators opposed this approach, asserting that balanced regulation would ensure foreign companies also adhere to the AI Act, fostering fair competition.

The AI Act defines artificial intelligence systems in alignment with internationally recognized criteria, drawing inspiration from OECD guidelines. It broadly categorizes AI systems as machine-based entities that deduce outputs, explicitly or implicitly, to influence physical or virtual environments based on received input for specific objectives. While encompassing various sectors, the AI Act exempts certain applications, such as military and defense, and distinguishes between the professional and personal use of AI.

The legislation introduces a dual classification system for AI systems, distinguishing between those posing limited and high risks. All AI systems are subject to fundamental transparency obligations, while high-risk AI systems, which have the potential for significant societal impact, are held to more comprehensive regulatory standards. The classification metrics, including computational power measured in FLOPS, are adjustable and subject to revision.

Prohibitions within the AI Act target practices considered harmful, such as manipulating individual cognition, the random collection of facial recognition data, and the deployment of social credit scores. Special obligations apply to general-purpose AI systems and foundation models, emphasizing risk management, red teaming, cybersecurity, and monitoring energy consumption.

The AI Act establishes stringent obligations for providers, requiring compliance with Union copyright laws, advanced technology adoption, and the preparation of comprehensive model training summaries. Mandatory registration in a European database and content disclosure may lead to legal challenges from rights holders.

For governance and compliance, the legislation establishes a European AI Office, scientific panel, and advisory forum. Sanctions for AI Act violations are determined based on companies' global turnover or a predetermined amount, with exceptions and reduced penalties for smaller businesses and startups. A transition period outlines the timeline for the applicability of the AI Act, with the final text expected in January 2024.

At Hypatos, we offer enterprise-ready AI agents for high document throughput and are aware of our responsibility. Protecting our customers is important to us, which is why we generally welcome the new regulations.

If you have any questions, our Legal team is at your disposal - legal@hypatos.ai.

We look forward to engaging in a dialogue with you.