Europe Establishes AI Regulations

Europe Sets Ground Rules for Artificial Intelligence with New AI Act

A Landmark Regulation

The European Union has taken a decisive step towards regulating artificial intelligence (AI) by passing the Artificial Intelligence Act (AI Act). This groundbreaking legislation, approved in 2024, creates a unified legal framework for AI use within EU borders, potentially affecting global tech companies with European customers.

European Artificial Intelligence Board: Enforcing Compliance

In an effort to promote cooperation and ensure adherence to the new rules, the AI Act establishes the European Artificial Intelligence Board. This body mirrors the EU’s approach to data protection, extending its reach to AI providers outside the EU that serve European users.

Scope and Exemptions of the AI Act

The AI Act addresses a wide array of AI applications across various sectors while exempting those used exclusively for military, national security, research, and non-professional purposes. Notably, the Act has been updated to cover the surge in general-purpose AI systems, like ChatGPT, with plans for stricter regulations on those with systemic implications.

Risk-Based Classification System

The legislation categorizes AI applications based on their potential harm, ranging from unacceptable to minimal risk, and imposes corresponding obligations. Unacceptable risk applications face a ban, while high-risk ones must meet stringent requirements. Limited-risk applications carry transparency obligations, and minimal-risk ones are left unregulated. General-purpose AI systems face transparency mandates, with further scrutiny for high-capacity models.

Enforcement and New Institutions

The AI Act calls for the creation of new institutions within the EU to implement and enforce its provisions. Member States are tasked with designating national competent authorities to oversee market compliance and conduct surveillance.

Market Entry and Conformity Assessments

To enter the EU market, AI systems must satisfy essential requirements, which will be further defined by European Standardisation Organisations. Conformity assessments, either self-conducted or by third parties, will ensure compliance, though concerns have been raised about the lack of independent mandatory evaluations for high-risk AI systems.

Legislative Journey and Timelines

The AI Act’s journey began with a white paper in 2020 and involved extensive debates and negotiations. The European Parliament passed it on March 13, 2024. It was then approved by the EU Council on May 21, 2024. The law will take effect after publication in the Official Journal, with staggered applicability timelines based on the AI application type.

Global Impact and Criticism

Experts suggest that while the AI Act is European in jurisdiction, its implications could be global, influencing companies aiming to enter the European market. However, Amnesty International and other organizations have criticized the Act for not fully banning real-time facial recognition and for potential human rights risks associated with AI technology exports.

Industry Reactions: A Mixed Bag

The AI Act has garnered mixed reactions from the tech community. While some startups appreciate the clarity it brings, others fear it may hinder their competitiveness. Tech watchdogs have pointed out loopholes that could benefit large tech companies, and advocacy group La Quadrature du Net has expressed concerns over the Act’s potential for social control and environmental harm due to self-regulation and exemptions.