EU agrees ‘historic’ rules to regulate AI

11 Dec 2023

Image: © rh2010/

Thierry Breton said the AI Act, set to take effect in 2025, is a ‘launchpad’ for EU start-ups and researchers to lead the global AI race.

The European Union has become the first political entity to agree on a set of rules to regulate artificial intelligence after members of the EU Parliament and Council reached an AI deal over the weekend after days of negotiation.

First proposed in early 2021, the AI Act was passed by the Parliament in June to establish a set of rules to categorise different AI systems based on the risks they pose, monitor their use and prohibit them from being misused within the EU.

Calling it a “historic moment”, Commission president Ursula von der Leyen said that the AI Act “transposes European values to a new era” and that it will foster “responsible innovation” in Europe.

“By guaranteeing the safety and fundamental rights of people and businesses, it will support the development, deployment and take-up of trustworthy AI in the EU,” she said. “Our AI Act will make a substantial contribution to the development of global rules and principles for human-centric AI.”

The new rules, now subject to formal approval from the Parliament and Council, will be applied directly in the same way across all member states on a risk-based approach involving four categories: minimal, high, unacceptable and specific transparency risk.

While minimal risk AI such as recommender systems or spam filters will get a “free pass” and have no obligations, high-risk systems involving critical infrastructure (water, electricity and education), law enforcement and certain biometric systems will be subject to strict rules.

AI systems considered a “clear threat” to the fundamental rights of EU citizens will be considered unacceptable and will be banned.

This includes AI that manipulates human behaviour to “circumvent” a user’s free will, such as toys using voice assistance that encourages dangerous behaviour in minors or systems that allow “social scoring” by governments or companies.

Predictive policing, emotion recognition systems at the workplace and real-time biometric identification for law enforcement will also be prohibited, with “narrow exceptions”.

Under the specific transparency risk category, the EU will require AI systems such as chatbots to disclose to humans that they are interacting with a machine.

This means that deep fakes will have to come with a label and users will have to be informed when biometric categorisation or emotion recognition systems are being used. AI-generated audio, video, images and text will have to be marked as such in a machine-readable format.

According to rules agreed over the weekend, the AI Act will require companies not complying with the rules to pay hefty fines up to €35m or 7pc of global turnover (whichever is higher).

“More proportionate caps are foreseen for administrative fines for SMEs and start-ups in case of infringements of the AI Act,” a statement by the Commission reads.

Even though an agreement has been reached, it will be a while before the rules are fully enforced. The AI Act is set to become applicable two years after its entry into force, the Commission said, except for some specific provisions. Prohibitions will apply after six months while the rules on general purpose AI will apply after a year.

“The AI Act is much more than a rulebook – it’s a launchpad for EU start-ups and researchers to lead the global AI race,” EU internal markets chief Thierry Breton posted on X.  “The best is yet to come!”

Last week, the European Digital SME Alliance raised concerns that potential changes to the AI Act, such as allowing Big Tech companies to self-regulate their foundation AI models, could shift responsibility to smaller businesses and create barriers to entry. It is not clear if this issue has been addressed.

10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.

Vish Gain is a journalist with Silicon Republic