EU member states have reached an agreement on important AI laws after they were passed conditionally in December. The text appears to be finalized and there are still a few rounds of voting planned.

The EU member states have agreed on the content of the new AI legislation. A conditional text was adopted for the first time in December, but still required further negotiations. Some countries, including France and Germany, wanted guarantees that the so-called AI law would not undermine innovation.

These guarantees now take the form of a series of innovation measures as well as a Artificial Intelligence Office Who has to take care of the implementation of the AI ​​law? This authority will also prevent these provisions from overlapping with other legislation.

Approved permit

After member states agree to the December rules, a committee in Parliament can approve the final text. After this approval, it is time for the actual approval by the plenary session. After this fourth approval, the AI ​​law was actually approved. Six months after this hurdle, the first rules to prevent the misuse of AI could come into force.

With the AI ​​law, the EU is taking a pioneering role. With these rules, the EU will be the first place in the world with clear AI laws.

AI is evolving so quickly that it has always lagged behind. Therefore, there is great concern about these rapidly growing AI developments. AI legislation has been in the works for more than two years, but final approval remains difficult. The most important hurdle was overcome in December: after another meeting of the European Parliament, an agreement was reached on December 8th in Brussels.

AI categories

The interpretation of the new legislation is based on categories. These are assigned based on how risky a particular AI application is. For example, simple spam filters or AI systems that give users tips are considered “minimal risk”.

AI systems used in the areas of justice, health, biometrics and water or energy management, on the other hand, fall under “high risk” and are therefore subject to stricter rules. Additionally, systems that manipulate human behavior are considered “unacceptable risk.”

Finally, there is a final category: “transparency risk”. Users need to be notified when they come into contact with AI. These include, for example, chatbots and deep fakes. The latter are AI-fake videos of existing people that look very realistic.

Violations are punished with a high fine

Violating these AI rules has serious consequences. The fines amount to a maximum of 35 million euros or 7 percent of global sales. This depends on the type of violation and the size of the company.

This article was originally published on December 11, 2023. The text has been updated with the latest information.