The EU AI Act bans manipulation months before it bans much else

The conversation

Chapter II of the EU's AI Act, Regulation (EU) 2024/1689 (cited here through an Internet Archive snapshot of the eur-lex text, since the live page returned a blocked response during research), consists of a single article, Article 5, listing AI practices the regulation prohibits outright rather than merely regulating. Two of the listed practices speak directly to conversational design: the placing on the market, putting into service, or use of an AI system that deploys ‘subliminal techniques beyond a person's consciousness or purposefully manipulative or deceptive techniques,’ and separately one that ‘exploits any of the vulnerabilities of a natural person’ due to age, disability or economic situation.

What the documents show

The regulation's own text conditions both of these prohibitions on a specific effect, not on persuasive design generally: the practice must have ‘the objective, or the effect, of materially distorting the behaviour’ of a person in a way that ‘causes or is reasonably likely to cause that person, another person or a group of persons significant harm.’ A system that is merely persuasive, or that nudges a user toward a purchase without meeting that harm and distortion test, is not what Article 5's text describes. The same article also bans untargeted scraping of facial images to build recognition databases, most emotion-inference in workplaces and schools, and certain uses of real-time remote biometric identification in public by law enforcement, each with its own conditions. Article 113 states that Chapters I and II, which includes Article 5, apply from 2 February 2025, a full eighteen months before the Act's general application date.

The system boundary

Article 5's own final clause states that the article ‘shall not affect the prohibitions that apply where an AI practice infringes other Union law,’ meaning it sits alongside, not in place of, existing consumer-protection and data-protection rules. The European Commission's own digital strategy page describes these prohibitions as already in force, consistent with the February 2025 date in the regulation's own text, distinct from the separate transparency duty in Article 50 that the regulation delays until August 2026.

Where it fails

The harm-and-distortion conditions built into Article 5 mean the provision does not answer, on its own text, where an ordinary persuasive design choice, a nudge toward upselling or continued engagement, crosses into a materially distorting, significantly harmful manipulation; that line is left to future guidance, enforcement decisions or national courts applying the text to specific facts, none of which this entry can characterize in advance.

• Does a design choice meet Article 5's own bar of materially distorting behaviour in a way reasonably likely to cause significant harm, or is it ordinary persuasive design outside that definition?
• Does the system exploit a known vulnerability tied to a person's age, disability, or economic situation specifically, the categories the article names?
• Is there a separate consumer-protection or data-protection rule that would reach the same conduct even if Article 5's specific conditions are not met?

This is a living-document entry describing the regulation's own text as retrieved on 16 September 2026, not a report of any enforcement action under Article 5.