The European Council adopts the AI ​​Act, what are its main measures to protect you from its abuses?

The European Union has just definitively adopted the AI ​​Act, the first global legislation aimed at regulating the development of artificial intelligence. An unprecedented, bold, but also highly anticipated legal framework to both stimulate AI innovation and protect the fundamental rights of users.

This is a world first that the European Union has just accomplished. After more than two years of intense negotiations, the European institutions definitively adopted the AI ​​Act after announcing it in December 2023. This pioneering legislation aims to regulate the development and use of artificial intelligence. .

Coincidentally with the calendar, Emmanuel Macron opened two days on May 21, 2024 dedicated to artificial intelligence, hoping that France would be a strong place for it. The European Council appears to have struck the delicate balance between encouraging technological innovation and ensuring respect for fundamental values ​​and rights. For their part, the United States is aiming for the same objective with its US AI Safety Institute Consortium (AISIC).

By classifying AI systems by level of risk, with proportionate rules, Europe intends to both stimulate the development of AI “ trustworthy » in its single market and protect its citizens against potential abuses and abuses.

Risk levels, a pioneering legal framework born from a marathon legislative process

Proposed in April 2021 by the European Commission, the AI ​​Act was eagerly awaited to fill a legal void in the face of the proliferation of AI systems, which are promising, but also give rise to serious concerns. After 18 months of legislative shuttling between European institutions, a political agreement was finally found in December 2022 on this founding text of an ecosystem of trust for AI in Europe.

Following a unique approach based on risks, the AI ​​Act establishes a framework grading obligations according to the level of threat to security and fundamental rights. Systems with unacceptable risk, such as biometric mass surveillance or social scoring, are thus simply prohibited. Strict rules of approval and human control then govern so-called high-risk uses in areas as sensitive as human resources, education, critical infrastructure, law enforcement, etc. Transparency and abuse monitoring obligations apply to lower-risk AI. Finally, a category of minimal risks remains almost free of constraints.

To guarantee the effectiveness of the system, the AI ​​Act also sets up new European governance of AI: a control office within the Commission, a council of independent scientific experts, an operational committee of Member States as well as as well as mechanisms for involving stakeholders and civil society.

Heavy obligations for so-called high-risk AI to protect users, but also safeguards to guarantee innovation

At the heart of the AI ​​Act, it is the category of high-risk AI systems which concentrates most of the new obligations. Use of AI in recruitment, education, essential public services, police and justice, credit contraction, critical infrastructure: a whole series of very widespread use cases will now have to meet drastic requirements. Transparency and traceability of development, effective human control, regular monitoring of biases and deviations, rigorous testing, third-party certification… Developers and users of these high-risk AIs will have heavy constraints to respect in order to access the European market.

These are strong measures to protect users from possible abuses that these tools unfortunately already allow, such as “Nudify” type applications which undress women using AI, or even deepfakes spreading false information. .

At the same time, the AI ​​Act also introduces safeguards to enable the development of responsible and trustworthy AI innovation. THE ” regulatory sandboxes » will thus provide a secure framework for testing and developing AI solutions before they are fully deployed. Regular monitoring is also planned to adjust the lists of prohibited or high-risk systems, based on technical developments.

Finally, a specific and lighter regime is planned for the very popular general AI such as chatbot or AI engine. If they do not present a major risk, these models will have some transparency obligations. But in the event of proven risks for society, stricter rules may be imposed on them. Has the new era of AI finally arrived?

Source : Council of the European Union