European Union approves AI Act

On the evening of December 8, 2023, and after three days of intense negotiations, European Union member States and the European Parliament reached a “political agreement” on the AI Act. “A historic moment! The EU becomes the first continent to establish clear rules for the use of AI,” applauded the Internal Market Commissioner, Thierry Breton, who is behind the bill. Technical work remains to finalize the bill before putting it to a vote.

Presented in April 2021, the AI Act aims to establish “requirements for AI according to its potential risk and impact levels.” It thus divides AI models into three categories according to risk: low, limited and high. For the first category, the only transparency requirement will be a disclaimer making it clear that the content is AI-generated.

“High-risk” AI models, however, which pose a “significant potential threat to health, security, fundamental rights, environment, democracy and rule of law,” will have to comply with more drastic requirements. An impact study on fundamental rights will be mandatory, as will maximum transparency on model training and uses.

The AI Act also outlines “unacceptable” uses of AI, which will be completely prohibited in Europe: real time biometric identification, social credit scoring, subliminal manipulation, predictive policing, use of human vulnerabilities and facial recognition databases. The bill makes exceptions for law enforcement, who will be able to use biometrics in real time for critical investigations.

Back in early December 2023, the bill seemed on thin ice due to a potential withdrawal from France and Germany, under pressure from their startups. The latter feared the AI Act would prevent the growth of new European players due to overly strict rules.

The compromise includes measures that support the European AI ecosystem and guarantee European businesses, particularly SMEs, access to AI models. Yet European tech remains wary, for the moment.