The European Parliament has voted to confirm its negotiating mandate for the AI Act, marking a major step towards a unified rulebook for artificial intelligence (AI) across the EU. The parliament supported an amended version of the proposal, which seeks to ensure that AI developed and used in Europe aligns with EU rights and values, including human oversight, safety, privacy, transparency, non-discrimination, and social and environmental well-being.
Vote on #AIAct:— Thierry Breton (@ThierryBreton) June 14, 2023
Thank you @EP_President Roberta Metsola and Rapporteurs @IoanDragosT & @brandobenifei for the very strong support of the European Parliament.
Our European democracy in action! 🇪🇺 pic.twitter.com/yIpkDz2a47
Notably, the changes backed by MEPs include a total ban on remote biometric surveillance and predictive policing, as well as a prohibition on untargeted scraping of facial images from the internet or CCTV footage to create facial recognition databases. Emotional recognition tech used by law enforcement, border agencies, workplaces, and educational institutions is also banned under the proposed rulebook.
The amendments expand the classification of high-risk AI systems to those that could significantly harm people’s health, safety, fundamental rights, or the environment, as well as AI systems used to influence voters and the outcome of elections. Large social media platforms using algorithms for content recommendation have also been added to the high-risk list.
The proposal includes obligations for makers of general-purpose AI models, like the technology underpinning OpenAI’s AI chatbot ChatGPT. These obligations require that such systems identify and mitigate risks before entering the market, apply transparency disclosures to AI-generated content, implement safeguards against illegal content generation, and publish detailed summaries of copyrighted information used to train their models.
OpenAI CEO Sam Altman has expressed concerns about these obligations, suggesting they may necessitate withdrawal of the service in the EU if compliance is not possible. However, today’s vote showed strong support among parliamentarians for the amended version of the draft legislation.
The next step involves discussions between the parliament and EU Member State governments. While there is hope that an agreement on the AI Act can be reached by the end of the year, the legislation will likely not come into effect before 2026. In the meantime, the EU is working on voluntary initiatives to encourage AI firms to self-regulate on safety.