EU Commission issues guidelines on prohibited AI practices under EU AI Act

An alert from Wilson Sonsini reports that the European Commission issued draft guidelines on February 4, 2025 explaining which AI practices the EU AI Act forbids. The guidelines carry no binding force but give examples to help businesses. The Commission has approved the draft and is expected to adopt it formally soon. The ban on prohibited practices, together with provisions on AI literacy, took effect on February 2, 2025.

Article 5 of the Act covers systems seen as posing unacceptable risk, and applies to those who offer such systems and those who use them. Firms outside the EU can be caught too.

On social scoring, the guidelines indicate that pricing insurance or judging creditworthiness could qualify when based on unrelated personal traits, while loyalty perks for reliable online shoppers and user ratings of drivers or hosts fall outside it. Manipulation and exploitation include AI in games that exploits children's vulnerabilities to drive compulsive play, and scams aimed at older people. Building face databases by untargeted scraping of images is banned, but not collecting voice samples or keeping face images that are not used to identify anyone. Emotion recognition at work or in education is generally barred, with exceptions for medical and safety uses, and is not covered in commercial settings. Biometric categorization by sensitive traits is prohibited, though purely technical uses such as a virtual try-on filter are not.

Providers must avoid releasing systems reasonably likely to be used for banned purposes, adopt safeguards, state the prohibitions in their terms and keep monitoring. Fines can reach EUR 35 million or seven percent of global annual turnover, and enforcement is expected from the later half of 2025.