Some AI regulations do not merely impose conditions on risky systems but ban certain uses entirely, treating them as crossing a line that no safeguards can justify. The clearest articulation is Article 5 of the EU AI Act, the “unacceptable risk” tier, whose prohibitions became applicable on 2 February 2025. The European Commission describes these as AI systems that pose “a clear threat to the safety, livelihoods and rights of people.”
The EU AI Act bans, among other things, AI that uses subliminal or deliberately manipulative techniques to materially distort behavior in harmful ways; systems that exploit vulnerabilities tied to age, disability, or socioeconomic situation; social scoring that evaluates people based on behavior or traits and leads to unfair treatment in unrelated contexts; predictive policing based solely on profiling a person; creating or expanding facial-recognition databases through untargeted scraping of images from the internet or CCTV; inferring emotions in workplaces and schools; biometric categorization that deduces sensitive characteristics such as race, political views, or sexual orientation; and, with narrow exceptions, real-time remote biometric identification in public spaces for law enforcement. Other jurisdictions have adopted similar lists; Texas’s 2025 AI law, for example, prohibits government social scoring and certain biometric and manipulative uses.
The concept of prohibited practices marks the boundary of what a society will not allow automation to do regardless of accuracy or efficiency. For businesses, the practical takeaway is that some AI products are simply off-limits in regulated markets, and identifying those red lines is the first compliance step before any risk assessment.