AI Act: EU adopts legal framework for artificial intelligence
After around three years, the process for legal regulation on Artificial Intelligence (AI) has been finalized with the adoption of the AI Regulation by the EU Parliament. The regulation is regarded as the world’s first comprehensive codification of legal issues relating to AI. The aim of the AI Act is to establish a cross-sectoral legal framework, whereby a risk-based approach is pursued. The higher the potential risks of an application, the higher the legal requirements shall be. Certain AI applications are also banned altogether. Regarding Intellectual Property Rights and Copyrights, the Regulation only makes marginal stipulations.
The AI Act contains inter alia the following provisions:
- Prohibited Practices: AI solutions with unacceptable risk are prohibited on the basis of the Regulation. These include, for example, applications for social scoring, i.e. the evaluation of human behaviour.
- High-Risk AI Systems: Permitted, but very heavily regulated, is the use of high-risk systems, which includes for example medical devices, banking products or critical infrastructures. In these cases, a comprehensive catalogue of obligations must be fulfilled. These include documentation and transparency obligations as well as security requirements.
- Limited Risk/Minimal or No Risk: AI systems with limited risk are, for example, those that are used for the purpose of customer service, such as chatbots. Transparency obligations apply to these. The lowest risk level is set at minimal or no risk AI applications, the use of which is free. These include systems that minimise production processes or spam filters.
- General Purpose AI: Specific rules apply to AI systems with a special focus. To give one example: For General-Purpose AI (GPAI) the regulation also provides for certain transparency and copyright compliance requirements. This applies, for example, to AI systems that generate texts or images like ChatGPT.
The directive still has to be confirmed by the Council. The AI Act will then enter into force, whereby the requirements – with few exceptions – will apply in the EU member states within two years. The adopted text of the regulation can be found here.