AI risk classification is the approach taken by European regulation, which does not impose the same requirements on all systems but grades them according to their potential impact on people. At one end are practices considered unacceptable and prohibited, followed by high-risk systems with the strictest obligations, then systems with limited risk, where the focus is mainly on transparency towards the user, and finally ordinary applications with minimal requirements. The first step for every company is to classify its systems, because everything else follows from that – the scope of documentation, the need for human oversight, and obligations towards suppliers. Classification is done according to the purpose of use, not the technology: the same model can be an ordinary application in marketing and a high-risk system in recruitment. Check the current wording and deadlines in the applicable regulation.
See also: High-risk AI system, Register of AI systems, Provider and deployer.
