EU Digital Fairness Act Targets Dark Patterns and AI Shopping Agents

Targeting engagement, not just deception

The draft would require that any user can switch off four types of features. Among them are notifications unrelated to what the user has done, which amount to spam, and chatbots that simulate a relationship to stop people from leaving.

None of these tactics deceives anyone, yet they are effective. Autoplaying the next video or a streak you hate to lose are legal and highly profitable product decisions. Until now the EU pursued deception as such; the draft begins to address the hook itself.

There is a catch: the EU is not banning these features, only requiring a switch to turn them off. Platforms know how few people dig into settings.

Rules for AI shopping agents

For the first time in European consumer law, the draft sets rules for AI agents that buy on a user's behalf. Users decide which purchases need their approval, and the agent must flag when a suggestion is sponsored.

It looks modest, but this is where manipulation is likely to move. When an agent shops, the trap will no longer be a grey button next to a green one, but a paid recommendation delivered in a neutral tone in a conversation nobody else can see. The EU wants it labelled before that market exists.

Penalties and scope

The text provides fines of up to 6% of global turnover, as under the DSA. The 2005 directive becomes a regulation applied uniformly in all 27 member states. The Commission could also extend the list of banned practices without going back to legislators, which is likely to worry the tech industry most.

Europe is very late on web traps seen since the early 2010s, but with AI agents it may, for once, be ahead of the problem.