After a final marathon negotiation round of almost 2 full days and nights, European lawmakers agreed on the long-anticipated AI Act. With this deal Europe becomes the first continent to set a comprehensive rulebook for AI that will serve as a blueprint for the world.

The AI Act is the first in the world to set specific rules for AI based on the protection of fundamental rights and European values. The rulebook takes a risk-based approach. The higher the risk that an AI system poses to health, safety or fundamental rights, the stricter the rules, with some AI practices even being banned entirely. “Today marks a true milestone for trustworthy AI. This law has been almost five years in the making, starting with the work of the High-Level Expert Group on AI and its Ethics Guidelines for Trustworthy AI”, says ALLAI President and Co-founder Catelijne Muller, who was member of the High-Level Expert Group on AI that developed the Guidelines. “I am happy that our work served as a strong basis for the AI Act.”

ALLAI has been closely involved in the lawmaking process ever since and has been advising the European Parliament’s co-rapporteur Brando Benifei and his team as well as several shadow rapporteurs on topics such as biometric recognition, general purpose AI (foundation models) and high-risk AI systems. “I am proud that quite a number of our recommendations were taken on board. This law will help improve the quality of AI and at the same time protect the fundamental rights of European citizens.”

Bans, high risk AI obligations and general-purpose AI

The text of the political deal that was reached tonight is not available yet, but from what we know now, the AI Act will contain more, and stronger bans than initially proposed by the European Commission. The European Parliament managed to negotiate additional bans on predictive policing, biometric categorisation, emotion recognition in education and the workplace and the online scraping of facial images. Moreover, the bans on AI-driven manipulation and exploitation of vulnerabilities and social scoring were strengthened. “We welcome the strong prohibitions”, says Catelijne. “These AI practices jeopardize virtually all fundamental rights and rightly have no place in Europe”. 

High risk AI systems must undergo mandatory fundamental rights impact assessments, including AI systems aimed at influencing the outcome of elections and voting behaviour as high risk. Obligations for general purpose AI (foundation models) are explicitly included in the act, which is important, particularly in light of the recent call by some Member States to leave them to self-regulation. “This will hopefully incentivise their makers to make a better product,” Catelijne argues. One that can be safely used in critical domains and actually make a difference for society.”

Breaks, seatbelts, and road signs

ALLAI has been at the forefront of advancing AI that aligns not only with our laws (new and existing), but also with ethical principles and social values and norms. “The AI Act is an important milestone. It provides the breaks, seatbelts and road signs that are much needed for AI at this moment”, says ALLAI’s co-founder Virginia Dignum. “But we cannot sit back and relax. AI is a socio-technical system that needs to adapt to the dynamics of society. This requires constant attention”.

What’s next

The final text of the AI Act will be officially adopted early 2024. After a gradual transition period the act will become law in all EU Member States mid 2024 (bans), early 2025 (GPAI) and early 2026 (full law). “The AI Act is complex and difficult to navigate, so public and private organisations need to start preparing now.” Catelijne concludes.