AI Act — A New Era for AI Regulation

On Wednesday 13th March MEPs finally approved the AI Act. It has been a long journey since its proposal on the 21st April 2021, due to its ambitious aim. To this day, the EU is the first entity in the World to regulate Artificial intelligence. But how will it change the lives of European citizens and companies?

The Act bans applications that exploit users’ vulnerabilities and threaten citizens’ rights. For instance, AI systems that can manipulate opinions or classify people by their social behaviour are indeed prohibited. Furthermore, the AI act’s fundamental goal of promoting trust towards AI systems is achieved through a risk-based approach, establishing obligations for applications according to their level of danger. Burdensome obligations, i.e. a conformity assessment, are provided for ‘high-risk’ AI systems. These include biometric categorisation systems based on sensitive characteristics and their expenditure in education, employment, essential private and public services, border management, justice and democratic processes. Therefore, the high-risk AI systems providers are called to fulfil disclosure and transparency obligations to users, ensure human oversight, accuracy and cybersecurity. As a result, users will be allowed to file complaints about their treatment and to receive explanations about decisions based on AI that affect their personal sphere.

However, even if this significant step forward in the European legal landscape aims to enhance innovation, so as to help Europe affirm its leadership in the technology field, several concerns relating to the overly restrictive approach are rising. It is feared that it could lead to limitation of progress and innovation. In fact, AI has a meaningful potential, and could bring consequential benefits in challenging fields (i.e., sanitation).

The European institutions in so have struck a balance between safety and progress, consistently with the approach of the already enforced GDPR.

What will the next steps be? The regulation is still subject to a final linguistic check and needs to be formally endorsed by the Council. The general-purpose AI rules will apply in May 2025, and the obligations for high-risk systems in three years from now. National authorities will play a crucial role in its implementation, with the support of the AI office in the European Commission.

CC: Gabriel Campisi and Francesca Posadinu