TikTok's Harmful Rabbit Holes Breach Digital Services Act

The European Commission has delivered a serious warning about the design of TikTok’s recommendation system. On 6 November 2026, it preliminarily found TikTok in breach of the Digital Services Act because the platform’s personalised For You feed can repeatedly direct users towards content about sadness, isolation, body image and other potentially harmful subjects. For children and vulnerable users, this raises questions that extend far beyond the amount of time they spend on a screen.

The investigation strikes at the heart of an increasingly important online safety issue. If a child watches distressing material, the algorithm may interpret that attention as interest and deliver more of the same, even when the child wants to see less. Parents need to understand these risks, while regulators must examine whether social media platforms can demonstrate that their safety protections actually work.

What the EU found about TikTok’s For You feed

TikTok recommends videos using engagement signals, including how long someone watches and which videos they like. According to the European Commission’s 6 November statement, repeated engagement with subjects such as weight loss, body image or mental health struggles can lead to increasingly similar recommendations. The Commission calls this pattern a harmful rabbit hole.

Crucially, the Commission’s preliminary findings concern two separate failures, inadequate risk assessment and inadequate risk mitigation. It considers that TikTok did not properly assess how the system might contribute to harmful recommendation cycles, or introduce sufficiently effective safeguards against them. These are regulatory allegations at this stage, not a final infringement decision.

Why watching something does not mean wanting more

Children can watch upsetting material for many reasons. A video might provoke fear, concern, curiosity or confusion, while a young person experiencing emotional distress might remain focused on content that reflects how they feel. An engagement based system can misread that attention as a request for further recommendations.

This distinction matters because a child does not have to search deliberately for harmful material to encounter it repeatedly. A seemingly ordinary video about exercise, appearance or wellbeing can become the entry point to a narrower stream of similar material. The risk lies not only in individual videos but also in their frequency, sequence and cumulative effect.

Research shows how quickly harmful recommendations can escalate

Independent researchers have previously documented the rabbit hole problem. In Amnesty International Ireland’s account of its 2023 technical investigation, test accounts designed to simulate 13 year olds received a high volume of mental health related and potentially harmful videos after several hours of engagement. Amnesty reported that almost one in two recommended videos fell into that category after five to six hours, around ten times the volume received by comparison accounts with no demonstrated interest in mental health.

That striking figure describes controlled research accounts, not a survey of all teenagers or a prediction of what every child will see. Nevertheless, it illustrates why child safety cannot depend on the assumption that a recommendation feed will naturally correct itself. Researchers must assess the patterns an algorithm produces, not only whether it removes individual prohibited posts.

The concern did not disappear with time. In October 2025 follow up research, Amnesty International reported that teenage test accounts which engaged with mental health related content encountered videos romanticising suicide or presenting information about methods within three to four hours.

A peer-reviewed study published on 28 October 2026 provides further context. Researchers at the University of Melbourne interviewed 25 TikTok users who reported current eating disorders and examined their experiences of algorithmically reinforced content. Participants described encountering eating disorder related recommendations after initially viewing seemingly ordinary health and fitness material, while attempts to avoid harmful content often proved difficult.

The Melbourne study is qualitative and cannot establish how commonly these experiences occur across TikTok’s entire user base. Even so, it helps explain why repeated exposure, social comparison and a sense of belonging within an online community can make disengagement more complicated than simply telling a young person to stop watching.

The Not Interested button may not interrupt a harmful feed

Perhaps the most concerning element for parents is the Commission’s assessment of the controls offered to users. Its preliminary findings say that selecting Not Interested may not sufficiently change the feed because viewing time can continue to outweigh someone’s explicit preference to see less of particular material. Content diversification measures may also fail to reduce repeated exposure adequately.

The Commission believes TikTok should give greater weight to explicit choices, make recommendation controls easier to find and use, regularly test its algorithm and stop using viewing time as a recommendation signal. These are examples of changes the Commission considers necessary at this stage, rather than measures already imposed by a final ruling.

Parental controls cannot carry the whole responsibility

Parents are frequently encouraged to activate screen time restrictions, family supervision and content controls. Those measures can still form part of a sensible safety plan, but they do not necessarily address the recommendations being generated by the platform itself. A parent may limit access while remaining unaware that a child encounters repeated distressing material during the time they are online.

In February 2026, the European Commission separately questioned the effectiveness of TikTok’s screen time and parental controls when examining its potentially addictive design. Its concerns included infinite scrolling, autoplay, notifications and the highly personalised recommendation system. The November findings add another dimension by challenging whether users can reliably steer the content they receive.

CODA has previously examined independent testing of social media protective features in Child Safety Features Failed Independent Tests. Together, these developments challenge the familiar assumption that the existence of a safety feature proves that it offers effective protection. Parents deserve clear evidence of what these tools can and cannot do.

Free platforms still have a commercial cost

Most families do not pay a subscription to use mainstream social media services. However, advertising-supported platforms depend on attracting attention and encouraging repeat visits, which creates a commercial incentive to keep users engaged. This does not establish that any company set out to harm children, but it does make independent scrutiny of product design essential.

When viewing time becomes a strong recommendation signal, a platform may continue to promote content that holds attention even when it causes distress. The central question is whether corporate incentives, safety requirements and the interests of children can be brought into genuine alignment. Voluntary assurances alone cannot settle that question.

Regulators are now scrutinising the design itself

Today’s announcement forms part of a wider regulatory challenge to social media business practices. In February 2026, the Commission issued preliminary findings about addictive design, while July 2026 findings questioned whether TikTok’s settings adequately protected minors whose accounts and content could become publicly accessible. The November announcement focuses more directly on the way personalised recommendations may reinforce harmful themes.

These proceedings show why online safety policy must look beyond removing obviously illegal content. A sequence of individually lawful videos can still produce serious concerns when an automated system repeatedly recommends them in a harmful context. The Commission is therefore testing whether TikTok has met its responsibilities under the Digital Services Act to identify and mitigate systemic risks.

What this means for Ireland

Ireland has an important role in this investigation because its Digital Services Coordinator, Coimisiún na Meán, assisted the European Commission. The Commission is leading these proceedings under the Digital Services Act, and its findings could influence wider expectations for how major platforms evaluate and manage recommendation risks. Irish parents and schools have a direct interest in whether those regulatory expectations produce meaningful changes.

TikTok has the right to inspect the case file and respond to the preliminary findings. If the Commission ultimately confirms non compliance, it may issue a decision carrying a fine of up to 6% of the provider’s total worldwide annual turnover. However, financial penalties are not the same thing as proven improvements to children’s everyday experiences online.

What parents can do now

CODA continues to support delaying children’s access to social media, particularly for children under 13, alongside a broader smartphone-free childhood approach and reduced recreational screen time. Parents should also remember that minimum account age rules and basic safety settings do not establish that a platform is developmentally appropriate or free from risk. Where older teenagers use social media, regular conversations about what appears in their feeds are as important as conversations about how long they spend online.

Ask whether particular subjects keep appearing, including videos about body image, dieting, loneliness or emotional distress. Explain that watching a video, even unintentionally, can influence future recommendations, and encourage teenagers to tell a trusted adult if content becomes upsetting or difficult to avoid. Review the platform’s recommendation controls together, while remaining aware of the limitations identified by the Commission.

If online content appears to be affecting a child’s wellbeing, respond with curiosity rather than punishment. A supportive conversation, appropriate professional advice when needed, and a reduction in exposure can help address immediate concerns. Families can also consult CODA’s TikTok parental controls guide and parent resources.

When the algorithm creates the risk, settings are not enough

Today’s findings raise a fundamental question about responsibility. If a platform’s design repeatedly exposes children to potentially harmful recommendations, parents should not have to compensate for that risk through imperfect controls or constant monitoring. Companies should demonstrate, through meaningful and independent testing, that their recommendation systems protect young users.

The Commission has not yet made a final finding of infringement, and TikTok must have a fair opportunity to respond. Nevertheless, the investigation reinforces a wider point for legislators, researchers, educators and parents: child safety requires enforceable design standards, transparent evidence and effective oversight, not merely the promise that users can change their settings.

References and further reading

About Children of the Digital Age

Children of the Digital Age provides online safety education and training for schools, parents and organisations. Our guides support practical conversations about online behaviour and the choices people make. However, education and app settings do not replace safeguarding or professional support.

Send a message

Contact Children of the Digital Age to enquire about online safety education, parent sessions or training for your school or organisation.

TikTok’s Harmful Rabbit Holes Breach Digital Services Act