Commission preliminarily finds TikTok may have under-assessed recommender risks
The case focuses on repeated harmful content, minors, viewing-time signals and user controls, with a possible penalty of up to 6% of global annual turnover.
The European Commission has preliminarily found that TikTok may have insufficiently assessed the systemic risks created by its “For You” recommendation feed, Poland’s Office of Electronic Communications (UKE) said on 9 November 2026. The concerns include excessive exposure to repetitive content that could affect users’ mental health and well-being, including by deepening sadness, social isolation, eating disorders and negative perceptions of body image.
For compliance teams, the issue is whether TikTok properly identified and mitigated systemic risks arising from the design of its recommender algorithm. The Commission questions the effectiveness of TikTok’s safeguards, including the “Not interested” control, and says a final infringement decision could carry a fine of up to 6% of the platform’s total annual worldwide turnover.
What the Commission questioned
TikTok’s feed selects personalised content using signals including videos watched and liked. The Commission says this may create a “rabbit hole effect”, in which the subject matter narrows progressively and users are repeatedly exposed to similar material.
The risk is considered particularly significant for minors and other people requiring special protection. The Commission’s preliminary view is that TikTok may have under-assessed the algorithm’s effect on user well-being and that its existing safeguards do not sufficiently limit how often similar content is recommended.
It also questions the weight given to viewing time as an indicator of user interest. In the Commission’s view, TikTok may rely too heavily on how long a user watches a video, rather than giving greater weight to preferences the user has expressly declared.
The findings draw on TikTok’s documentation and internal data, risk-assessment reports, scientific research and opinions from technology, psychology and psychiatry experts. The Irish Digital Services Coordinator supported the Commission’s work.
The controls now in focus
The Commission identifies four possible directions for change:
- giving greater weight to users’ directly expressed preferences;
- reducing the importance of viewing time in recommendations;
- improving the accessibility and effectiveness of controls over displayed material; and
- testing the algorithm regularly to identify risks and assess whether safeguards work.
Those are proposed directions at the preliminary stage, not requirements imposed by a final decision.
In Atlas’s view, the supervisory signal is significant because it reaches the design choices that determine repeated exposure, not only content moderation, advertising transparency or user complaint handling. Platforms should be able to show how their risk assessments test recommendation patterns, vulnerable-user impacts and the effectiveness of user controls, rather than treating those controls as effective merely because they exist.
The case forms part of a wider Commission proceeding on TikTok’s compliance with the Digital Services Act. That proceeding has also examined the potentially addictive nature of platform features, protection of minors, researchers’ access to public data and advertising transparency.
TikTok’s rights of defence
The Commission’s findings are expressly preliminary and do not determine the outcome. TikTok may inspect the materials gathered by the Commission and submit its position under its right of defence. Consultations with the European Board for Digital Services are also foreseen.
If the concerns are ultimately confirmed, the Commission may adopt a decision finding an infringement of the DSA and impose the financial penalty.
Permanent link to this Atlas analysis
Related Atlas analysis
Continue with coverage connected by market, topic and operating context.
- PolandCommission proposes election safeguards for VLOPs, including AI deepfake labelsThe draft also calls for pre-election recommender testing, six-month change records and public risk reports within 30 days of each vote.
- BrazilBrazil SPA lets closing betting operators tap reserves for bettor refundsOperators need SPA approval, individualised beneficiary lists and administrator declarations before reserve funds can reach Caixa.
- GreeceEBA cuts 2027 EU-wide stress test data by 55%The timetable starts in late January, with a first submission due by late April and a separate climate-risk module outside the main results.
Continue with Atlas
Move from this development into the relevant research, comparison and workflow.