In a historic move, the European Artificial Intelligence Act (AI Act), the world’s first comprehensive regulation on artificial intelligence, has officially entered into force. Designed to ensure that AI developed and used within the European Union is trustworthy and aligned with the protection of fundamental rights, the AI Act aims to establish a harmonized internal market for AI. This regulation is expected to encourage the adoption of AI technology while fostering a supportive environment for innovation and investment across the EU.
The AI Act introduces a forward-looking definition of AI based on a product safety and risk-based approach. It categorizes AI systems into several risk levels, each with specific obligations. Minimal risk AI systems, such as AI-enabled recommender systems and spam filters, face no obligations under the Act. Specific transparency risk AI systems, including chatbots and deep fakes, must disclose their nature to users. High-risk AI systems, like those used in recruitment or loan assessments, are required to meet stringent requirements, including risk mitigation and human oversight. Finally, AI systems deemed to pose an unacceptable risk, such as those enabling social scoring or predictive policing, will be banned outright.
The regulation also addresses general-purpose AI models, which are versatile AI systems capable of performing a wide range of tasks. The AI Act ensures transparency throughout the value chain and addresses potential systemic risks associated with these advanced models. Member States have until August 2, 2025, to designate national competent authorities responsible for overseeing the application of AI rules. The Commission’s AI Office will play a central role in implementing and enforcing the Act, supported by advisory bodies to ensure uniform application and provide technical advice.
Key Points of the AI Act:
- Minimal Risk: AI systems like recommender systems and spam filters face no obligations but can adopt voluntary codes of conduct.
- Specific Transparency Risk: AI systems such as chatbots and deep fakes must clearly disclose their nature to users, with content marked as artificially generated.
- High Risk: AI systems used in critical areas like recruitment and loan assessments must comply with strict requirements, including human oversight and high data quality standards.
- Unacceptable Risk: AI applications posing clear threats to fundamental rights, such as social scoring and certain uses of biometric systems, will be banned.
- General-Purpose AI Models: Transparency and systemic risk management are mandated for versatile AI models used across various applications.
Application and Enforcement:
- Designation of Authorities: Member States must appoint national competent authorities by August 2, 2025.
- AI Office: The central body for implementation and enforcement at the EU level.
- Advisory Bodies: The European Artificial Intelligence Board, a scientific panel of independent experts, and an advisory forum will support implementation and enforcement.
- Penalties: Companies can face fines of up to 7% of global annual turnover for violating banned AI applications, with lower fines for other violations.
Next Steps:
- Implementation Timeline: The majority of rules will apply from August 2, 2026, with some prohibitions taking effect after six months, and rules for General-Purpose AI models applying after 12 months.
- AI Pact: A voluntary initiative for early adoption of key obligations.
- Guidelines and Standards: The Commission is developing guidelines and co-regulatory instruments to facilitate implementation, with a call for participation in drawing up the first general-purpose AI Code of Practice.
Background:
- Political Agreement: Reached on December 9, 2023. Support for Startups and SMEs: Measures launched on January 24, 2024. AI Office Unveiled: On May 29, 2024.
- EuroHPC JU Regulation: Amended on July 9, 2024, to enable AI-supercomputers for training General Purpose AI models.
- Ongoing Research: Conducted by the Joint Research Centre (JRC) to shape and implement EU AI policies effectively.