The AI Act, the world’s pioneering attempt to comprehensively regulate rapidly advancing technology in an ethical manner, has reached a preliminary agreement between member states and the European Parliament.
This breakthrough occurred on Friday night following extensive negotiations throughout the day, capping a 35-hour effort that spanned from Wednesday to Thursday afternoon. The agreement holds significance amidst fervent lobbying from tech companies, stern warnings from civil society, and intense media scrutiny, acknowledging the potential global impact of the Brussels legislation.
“Historic! The EU emerges as the first continent to establish clear rules for AI usage,” exclaimed Thierry Breton, the European Commissioner for the internal market, actively participating in the debate. According to Breton, the AI Act serves not just as a rulebook but as a launchpad for EU startups and researchers to lead the global AI race.
The negotiations involved a rigorous exchange between governments and lawmakers, addressing complex technical issues. After reaching a tentative compromise on Thursday regarding foundation models powering chatbots like OpenAI’s ChatGPT, Friday’s discussions focused on the use of real-time biometrics, including facial recognition, in public spaces.
The central debate revolved around whether state authorities should be permitted to deploy AI-powered biometric systems capable of identifying and categorizing individuals based on sensitive characteristics. While MEPs considered these practices “intrusive and discriminatory” and proposed a ban across EU territory, member states argued for exceptions, especially in law enforcement.
The clash between national security and fundamental rights dominated Friday’s discussions. Spain, holding the Council’s rotating presidency, played a crucial role in representing member states. The negotiations experienced a prolonged recess for lawmakers to discuss the demands presented by the Spanish presidency. Scholars and activists, meanwhile, used social media to urge MEPs to resist law enforcement exemptions.
Ultimately, the Parliament conceded and agreed to “strict conditions” allowing authorities to use real-time biometrics for specific purposes, such as searching for abduction victims, preventing terrorist threats, and locating individuals suspected of serious crimes. However, biometric categorization based on sensitive characteristics and other prohibited uses were established.
According to Brando Benifei, a socialist MEP, and co-rapporteur, the ban on biometric identification was a hard-fought achievement, representing a robust regime against undesirable uses of AI in the Union and the market, as emphasized by Dragoș Tudorache, the other co-rapporteur.
Given the complexity of the issue, the compromise from these prolonged talks is likely to undergo further refinement in the coming days. Carme Artigas, Spain’s secretary of state for digitalization and AI, emphasized that every article had been agreed upon without leaving any “loopholes” to circumvent human rights.
Once the legal text, spanning hundreds of pages in articles and annexes, is rewritten, it will undergo a new vote in the European Parliament and approval from the Council. The final votes are anticipated in early 2024, followed by a gradual implementation of the law.
The AI Act, introduced in April 2021, is a groundbreaking effort to ensure the ethical development of transformative technology in the 21st century. It proposes a risk-based approach, categorizing AI-powered products into minimal, limited, high, and unacceptable risk categories. The law sets rules for transparency obligations, conformity assessments, and registration for high-risk AI products, with a ban on those deemed unacceptable.
Under the preliminary deal, companies violating the rules will face fines of up to €35 million or 7% of global turnover. The compromise addresses concerns raised after the introduction of foundation models, such as ChatGPT, in 2022, ensuring transparency requirements for chatbots.
Ursula von der Leyen, the president of the European Commission, hailed the AI Act as a global first, providing a unique legal framework for AI development, trust, and ensuring the safety and fundamental rights of people and businesses.
