EU finds TikTok's For You feed breached the DSA — 6% fine at stake

The European Commission has spent two and a half years building this case, and on Friday it named the defendant precisely: not the videos TikTok hosts, but the machine that picks the next one.

The European Commission has preliminarily found TikTok in breach of the Digital Services Act over the "rabbit hole effect" of its recommender system, the Commission announced on Friday. The finding covers two duties — assessing the risk and mitigating it — and concludes TikTok failed at both. If it survives TikTok's right of defence and becomes a non-compliance decision, the fine can reach 6% of the company's total worldwide annual turnover, a number nobody outside ByteDance can calculate, because neither TikTok nor its parent publishes global revenue.

The mechanism the Commission describes is unglamorous, and that is the point. The For You feed reads engagement signals — what you watch, what you like — to decide what comes next. When a user repeatedly watches or likes content about weight loss or body image, "the system interprets these interactions as positive signs of interest and recommends more of the same," the Commission writes. Repeated exposure, in its account, produces "increased feelings of sadness or anxiety, body dissatisfaction, lower self-esteem, or worsening of existing eating disorders and mental conditions."

TikTok has built tools to break that loop. On the Commission's reading, none of them work. Measures to recommend diversified content "do not sufficiently reduce how often similar content is recommended." The "not interested" button "may not be enough to change the content feed," because the algorithm "prioritises showing users content based on viewing time, over their choice to see less of certain content." Officials told reporters that feed-refresh controls help only briefly and that "not interested" stops working after a while, according to The Record's account of the briefing.

So the Commission has asked for a redesign, and the list reads like a spec for a different ranking objective: rely more on what users explicitly say they want, disable viewing time as a recommendation signal, make the controls easier to find and use, and test the algorithm regularly. That is the closest a major regulator has come to ruling that engagement optimisation is itself the violation.

What makes the case unusual is what the Commission deliberately left out. Officials said investigators avoided illegal content and content that breaks TikTok's own rules — the "least harmful" category, one called it — and focused instead on material that looks innocuous one clip at a time: the "what I eat in a day" format, "SkinnyTok." The harm, in this theory, is repetition rather than any single post. "If you get a chain of hundreds or even thousands of videos of the same kind of content, this drives people in what we call a rabbit hole, and this has a negative impact on people's mental well-being," one official said.

European officials described the findings as the first time EU regulators have directly challenged the engagement-based recommendation system of a major platform, and, to their knowledge, the first enforcement action of its kind anywhere. One framed the case as "addressing a business model" rather than a moderation failure. The evidence base is correspondingly architectural: TikTok's risk assessment reports, internal data and documents, its answers to multiple requests for information, the published scientific literature, and interviews with recommender-system, psychology and psychiatry experts. Ireland's media regulator, Coimisiún na Meán, assisted. Officials said they analysed roughly 20,000 videos with psychologists, psychiatrists and system-design specialists, and that the testing "didn't really need access to the exact algorithm."

That focus connects to an uglier thread. Friday's file lands after US senators spent the autumn pressing TikTok over an internal experiment that withheld a mental-health safeguard from about 10% of its US users, one of whom was 16-year-old Chase Nasca, who died by suicide in 2022. We covered the letters and the experiment in August — TikTok's safety experiment: senators demand answers.

TikTok's response was blunt. A spokesperson said the Commission "has made a series of statements that are fundamentally inaccurate and distort our platform's architecture and our commitment to user safety," adding that the company has "a strong record of protecting our community while giving users direct and intuitive control over their content experience." TikTok can now read the file and reply in writing while the European Board for Digital Services is consulted.

This is the third set of preliminary findings against TikTok under the same proceedings: addictive design in February, protection of minors in July, and now the recommender itself. The file also includes researcher data access, where findings landed in October 2025, while the advertising-transparency strand closed in December 2025 through binding commitments. Regulators drew a line between Friday's action and TikTok's September settlement with Alabama — at least $100 million, plus time limits and other restrictions for teen users — saying the state deal "does not go as deep" into the recommendation system. Meta was told in July to change the addictive design of Facebook and Instagram, and officials said the Commission is "looking more widely beyond TikTok" without naming targets.

The broader significance is that recommender systems are the largest machine-learning systems most people actually touch, and until now they have been governed through transparency: publish your risk assessment, document your choices. This case goes after the objective function. The Commission also described it as a "scene setter" for the KIDS Act proposed in September, which would bar platforms from relying solely on engagement signals when recommending content to children — a provision officials said was directly inspired by what they found inside TikTok.

Three caveats are worth holding onto. The findings are preliminary, TikTok has not yet answered them on the merits, and the Commission has never actually collected a DSA fine of this size — the February addictive-design findings remain unresolved nine months later, which tells you the pipeline, not the penalty, is the binding constraint. The causal claim also rests on internal documents and observational analysis, not a controlled trial, and "rely more on explicit preferences" is a direction rather than a measurable standard; a company can comply on paper and change little in the feed. And the remedy the Commission wants — stop weighting watch time — is the part of the product that makes the app worth billions, which is why this will be litigated harder than any content-moderation case Brussels has brought.

What to watch: whether TikTok's written reply lands before the Board for Digital Services weighs in, whether the first actual non-compliance decision arrives before the KIDS Act does, and whether the Meta file gets the same treatment. The legal precedent matters more than the fine: if a regulator can order a platform to stop optimising for watch time, every feed, Shorts queue and autoplay next-up is suddenly a regulated artefact.

If the harm only appears across a thousand legal videos, does the responsibility sit with the person scrolling or the system that queued them? Write to us via our contact page.