Digital Fairness Act puts dark patterns on a blacklist

The European Commission presented the Digital Fairness Act this week. It bans dark patterns and adds subscription rules. What happens when an AI system learns these tricks by itself, and what part do users play?
AI generated image - A laptop on a desk showing a subscription cancel screen with a small grey cancel link, illustrating dark patterns under the Digital Fairness Act

A cancel button that is harder to find than the sign-up button is no accident. The European Commission wants to make tricks like this illegal.

The Commission presented the Digital Fairness Act on 11 November 2026, together with a revision of the Consumer Protection Cooperation Regulation. This is a proposal. Parliament and Council still have to agree on the text, so details can change. The direction, however, is clear.

What the Digital Fairness Act changes

The Digital Fairness Act turns the Unfair Commercial Practices Directive from 2005 into a regulation. A regulation applies directly in all 27 member states, so there is no national version to wait for. And where the Digital Services Act covers online platforms only, the Digital Fairness Act applies to every business that sells to consumers.

The penalties are serious. In large-scale cases that affect consumers across the Single Market, the Commission itself can investigate and fine a company up to 6% of its worldwide annual turnover. National consumer authorities keep their role and get stronger enforcement powers.

Nothing applies yet. The Commission gives no date. Before the proposal came out, law firm CMS did not expect the Digital Fairness Act to be in force before the end of 2028 at the earliest. So this will take a while.

Dark patterns on a blacklist

The core of the Digital Fairness Act is an updated blacklist of practices that are banned in all circumstances. An authority does not have to prove harm case by case. A draft that leaked two days earlier listed ten such dark patterns. The Commission’s own questions and answers give examples:

  • fake scarcity, such as a false claim that an item is nearly sold out
  • shaming consumers for a choice the business does not like, such as a “We’re sad to see you go” message when they unsubscribe
  • defaults the consumer has to undo, such as pre-ticked boxes or items added to the basket automatically

Anyone who has tried to cancel a gym membership will recognise these.

Subscriptions get their own rules

For digital contracts the Digital Fairness Act requires a simple, unambiguous way to cancel. A free trial may only turn into a paid contract when the consumer actively confirms. Consumers get clearer information before a free trial ends or a subscription renews automatically. And when a problem in an ongoing contract cannot be solved any other way, they can reach a human at customer service.

I like most of it. That we apparently need a regulation to say you should be able to leave as easily as you joined tells you something about how some companies treat their customers.

Dark patterns, AI and the Digital Fairness Act

This is where the Digital Fairness Act gets more interesting than a list of banned tricks.

The blacklist describes patterns that people designed. A designer chose to grey out the “no” button. A product team decided to ask a third time. Those choices can be found, named and banned.

AI changes how such choices come about. More and more interfaces are tuned by systems that learn what makes people click, stay and pay. Give a system the goal of higher conversion and enough room to test, and it can arrive at a buried cancel link or a guilt-laden “no thanks” by itself. Nobody designed the pattern on purpose. The system found that it works.

Who is responsible when the system found it

For the law this makes little difference. The blacklist judges the result on the screen, not the intent behind it. A company that lets an optimisation system run cannot point at the system when a dark pattern appears. The practice is unfair, whoever or whatever chose it.

For organisations it makes a real difference. The risk no longer sits only in the design review. It sits in the objective you give a system and in how often you look at what it produces. An AI that optimises for retention needs someone checking what it does with the cancel flow.

Autoplay, AI companions and AI agents

The Digital Fairness Act also gives consumers more control over design that encourages excessive or compulsive use. They will be able to switch off features such as video autoplay and notifications that have nothing to do with what they are doing on the service.

The leaked draft also covered chatbot features that simulate a relationship to keep users engaged. The Commission’s summary does not mention those under the Digital Fairness Act. It places AI companions and chatbots under the safety-by-design rules of the proposed EU KIDS Act, which protects children. For consumers of all ages, the Digital Fairness Act adds a different AI rule: clear rights when an AI agent buys products on their behalf.

The Digital Fairness Act and the part users play

Here is where I part company with some of the commentary. Regulation sets a floor. It does not replace attention.

Treating every consumer as someone who cannot read a renewal date does them no favours either. Above the floor the Digital Fairness Act builds, it still pays to read what you agree to. Note when a subscription renews. Notice when the “no” option is small, grey and tucked into a corner.

My suggestion is to treat friction as information. When cancelling takes four screens and two offers, that tells you how the company sees you. When a chatbot seems a little too keen to keep the conversation going, ask what it is optimised for.

That is not an argument against the regulation. A blacklist protects the people who do not have the time or the knowledge to spot these tricks. It is an argument against assuming the regulation does all the work.

What organisations can do now

Nothing in the Digital Fairness Act applies yet. That gives organisations time to look at their own interfaces before a regulator does. Three questions cover most of the ground. Is cancelling as easy as signing up? Is the price, including renewal terms, visible before the customer commits? And if an AI system tunes your interface, does anyone check what it has learned to do?

I will come back to this as Parliament and Council work through the text. If you want to follow this and other EU rules that shape how you build and buy AI, a free Future Prep account is the easiest place to start.

Bas Hennis

Future Prep helps organizations prepare for the impact of AI and emerging technologies. We provide hands-on training, strategic advice, and smart tools for the responsible use of AI, governance, and digital resilience.LinkedIn

Newsletter
[sibwp_form id="1"]
Related Blogs
LATEST NEWS

AI governance is not a future problem

Regulation is already in effect. Your competitors are already building internal capability. The gap between ‘we are aware of AI’ and ‘we have operational control’ is closing, and it closes faster with a structured framework.

 

Book a 30-minute discovery call. No obligation. We will assess where your organisation stands and what a realistic starting point looks like.

No sales pressure. No jargon. Just a structured conversation about your organisation's AI readiness.

Scroll to Top