The European Union hopes that creating strict artificial intelligence regulations early in the technology's development will help address risks in a timely manner and help shape an international agenda to regulate artificial intelligence. Systems intended for use in "high risk" situations listed in the law's appendices must meet various standards, including transparency, accuracy, cyber security and training data quality. Some uses, such as Chinese style social credit scoring, are banned outright.

High risk systems must obtain certification from approved organizations before they can be placed on the EU market. A new "AI Office" will oversee implementation at the EU level. There are also more basic rules for "general purpose" systems that are used in different situations   some high risk, others not. For example, providers of such systems may be required to keep certain technical documents available for audit.

Droit & intelligence artificielle (IA) : En quatre jours, adoption de deux textes européens inédits.

1) Le règlement européen (UE) sur l'IA (https://t.co/gEprdST2Wv)

2) La Convention-cadre (@Coe) sur l’IA & les droits de l’homme (https://t.co/KWskSkFBaJ)

Une première mondiale. pic.twitter.com/ID84xxjN1D

— Nicolas Hervieu (@N_Hervieu) May 21, 2024

However, providers of particularly powerful general purpose AI systems must inform the European Commission if the system has certain technical capabilities. If the provider cannot demonstrate that their system does not pose a serious risk, the commission may recognize it as a "general purpose AI model with systemic risk," after which stricter risk mitigation rules apply. Meanwhile, AI generated content such as images, sound or text needs to be marked to protect against misleading deepfakes. The European Commission proposed a first draft of artificial intelligence legislation in April 2021, releasing a white paper in February 2020 that outlined a risk based approach. The European Parliament has pushed for tougher rules   such as a ban on police using real time facial recognition on live CCTV feeds.

However, EU member states have been reluctant to impose more restrictions on law enforcement and border security and fear that excessive red tape could harm economic competitiveness. After several rounds of late night negotiations, the parliament and member states' negotiators finally reached a compromise in December. The final law imposes a general ban on real time facial recognition on CCTV but includes exceptions for law enforcement uses such as finding missing persons or kidnapping victims, preventing human trafficking, or finding suspects in serious criminal cases. The law is now finalized by a unanimous vote among ministers, must be signed by the presidents of the EU legislature and then published in the EU statute book. It becomes law 20 days after that, but most of its provisions don't go into effect for two years after that.