Article 5 is a defined list of prohibited AI practices, not a broad statement that every risky or inaccurate AI model is illegal. Context, intended conduct and timing are central.
Manipulation and exploitation have elements
Certain systems using subliminal, purposefully manipulative or deceptive techniques are prohibited where the legal tests concerning material distortion of behavior and significant harm are met. Separate restrictions concern exploitation of vulnerability due to age, disability or social/economic circumstances. “Persuasive advertising” and “prohibited manipulation” are not interchangeable labels.
Social scoring and predictive crime risk
The article targets social scoring that leads to disfavored treatment outside context or unjustified/disproportionate treatment, and specified crime-risk predictions relying solely on profiling or personality traits. A trained analyst reviewing objective, verifiable facts is a different factual pattern from treating a model-derived personality score as evidence of criminality.
Facial databases, biometrics and workplace emotion
Untargeted scraping of internet or CCTV facial images for face-recognition databases is covered. Certain biometric categorisation and workplace/school emotion inference are separately restricted; specific exceptions and definitions must be checked. A system may be prohibited even if it was marketed as “safety analytics.”
Beware the 2 December 2026 change
The consolidated Article 5 text includes additional restrictions concerning AI-generated intimate material without defined consent and certain child sexual-abuse material. The Service Desk identifies 2 December 2026 for those additions. On the stated cutoff date, do not misrepresent a future application date as if it had already arrived.
Operational review
Create a prohibited-use register and require human legal approval before launching surveillance, profiling, high-pressure persuasion or synthetic intimate-media products. Capture system purpose, available functions, foreseeably reproducible misuse, safeguards, deployer activities and Member State context. Read the binding consolidated text rather than relying solely on explanatory summaries.
Sources & Further Reading
- EU AI Act — Article 5 prohibited practices (consolidated 27 July 2026) — European Commission AI Act Service Desk.
- EU AI Act — consolidated version of 27 July 2026 — EUR-Lex.
Jurisdiction/scope: European Union; Article 5 consolidated July 2026. This is educational information, not legal, medical or other professional advice. No hands-on product testing is claimed.