On 13 March 2024, the European Parliament (Parliament) approved the EU Artificial Intelligence (AI) Act. The agreed regulation is expected to be adopted in April 2024. The regulation will enter into force in all EU member states 20 days after the publication in the Official Journal, which is expected in May or June this year.

The EU AI act is the first of its kind and aims to safeguard EU values and core principles like strengthening the internal market, rule of law and more, but also to promote trustworthy AI, amongst other ambitions.

The regulation categorizes AI-systems in four risk levels (unacceptable, high, limited/transparency and minimal).

• Unacceptable risks are prohibited use of AI-systems. For examples, social scoring by the state or real-time remote biometric identification of an individual in public as a part of law enforcement. There are exceptions to the latter.
• High-risk are AI-systems for areas like education, law enforcement, banking, employment, critical infrastructure, healthcare and more (see the annex to the AI act). For a high-risk AI-system to be put on the EU-market it needs to be pre-assessed and then registered in an EU database. High-risk AI-systems will be allowed following the regulation, but subject to restrictions.
• Limited/transparency risks are AI-systems like chatbots where users need to be informed when they are interacting with AI. In addition, AI-generated content needs to be identifiable meaning, for example, a text generated with AI for the public interest needs to be identified as being created as such.
• Minimal risks are free to use and is, for example AI-systems for spam filters.

Obligations arise for natural or legal persons, public authorities, agencies, or other bodies who are providers and/or deployers of AI-systems. The scope also extends to others including importers and distributors. The regulation also applies, in certain situations, to subjects in third countries, meaning outside the EU. One example is when providers or deployers are in a third country, but where the output of the AI-system is used in the EU.

The regulation will take effect gradually starting with the unacceptable risk six months after the regulation has entered into force. Provisions concerning high-risk AI-systems will be applicable 36 months after the regulation has entered into force.

Only time can tell whether the EU AI act will serve as blueprint for other jurisdictions to regulate AI.

TAGS

• AI
• Artificial Intelligence
• European Union