Leaked EU digital fairness draft puts AI shopping agents and engagement tactics in focus

AI shopping agents and chatbot features designed to prolong engagement could face new consumer protections under a leaked European Commission draft reported by Euronews on November 9. The reported Digital Fairness Act text would let users choose which purchases require approval and require disclosure of sponsored suggestions. It would also let people disable relationship-simulating chatbot features that keep them engaged. Euronews report
The draft is not adopted legislation. Euronews said the Commission declined to comment on the leak, with a formal proposal expected November 11. Its contents and timetable should therefore be treated as provisional.
AI enters a wider consumer-protection debate
The Commission's official consultation, opened in July 2025, identified manipulative interfaces, addictive product design, misleading influencer marketing and unfair personalization as problems the planned act would address. It specifically highlighted commercial exploitation of vulnerabilities and protection of minors. The consultation also framed simpler business rules and effective consumer protection as complementary goals. Commission consultation
That background matters because an AI purchase assistant is both a software system and a commercial interface. It may interpret an instruction, rank products and initiate a transaction. Each step can affect what the customer understands and controls, even if the underlying model answers questions fluently.
The design issue becomes especially clear when a recommendation is sponsored. A shopper needs to understand whether the assistant is selecting an option because it meets the request or because a commercial relationship affects the ranking. Disclosure must be understandable at the moment it can influence a decision.
A broader blacklist is reported
Euronews says the draft would prohibit ten manipulative design practices, including obstructive cancellation, concealed renewal terms, preselected extras and misleading scarcity claims. Other reported measures include active consent before free trials become paid subscriptions and penalties reaching 6% of global turnover. The text would still require agreement from Parliament and member states, with application proposed 12 months after entry into force. Draft reporting
These provisions point toward a practical distinction between permission to help and permission to spend. A useful assistant can narrow choices without receiving blanket authority to buy. Interfaces can make that boundary visible through clear approval settings, receipts and an accessible way to cancel or revise instructions.
What builders can examine now
The sensible preparation is a review of existing product behavior, rather than assuming the leaked language will become law unchanged. Teams can inventory sponsored placements, purchase permissions, renewal flows and engagement prompts. They can test whether a person understands those controls without reading lengthy explanations.
The same review should distinguish a user-requested reminder from a prompt intended primarily to bring the user back. Both may appear as notifications, but their purpose and the user's expectations can differ substantially.
The leaked text should be compared with the Commission's published proposal and any later amendments. The November 9 report offers a direction of travel rather than a final compliance checklist: AI assistants are being examined not only for what they generate, but for how their interfaces steer consumer decisions.
Sources and citations
Revision history
Published .