What it asks for. Article 5 lists AI practices that are banned in the EU because they pose unacceptable risk. These include manipulative or deceptive techniques that cause significant harm, exploiting vulnerabilities of specific groups, certain social scoring, predicting criminal offences based solely on profiling, untargeted scraping to build facial recognition databases, emotion recognition in workplaces and education with limited exceptions, certain biometric categorisation, and, with narrow exceptions, real time remote biometric identification in public spaces for law enforcement.
In practice. Organisations screen existing and planned AI uses against the prohibitions, document the assessment and stop or redesign anything that falls within them. Procurement checks should cover vendor functionality too. Evidence includes screening records and decisions.
Common gaps. Emotion recognition and biometric features in HR or security tools are easy to miss. Screening is often done once rather than as part of every new AI use.