In February of 2026, the European Commission preliminarily found TikTok in breach of the Digital Services Act (DSA) over what it called the platform’s “addictive design”, particularly as it relates to minors. It reached a structurally identical conclusion against Meta in July, citing infinite scroll, autoplay, push notifications, and personalised recommender systems as evidence that platforms’ design features “fuel the user’s urge to keep scrolling and shift the brain into ‘autopilot mode’”.
Addiction has recently become a buzzword in the discourse on platform governance. Recitals 81 and 83 of the DSA mention “addictive behaviour’’ and “behavioural addictions” resulting from the online interface design of “very large online platforms” (VLOPs) as an example of “systemic risks” to public health. However, the scientific literature holds no clear consensus on the determinants of behavioural addiction (Goldberg, 2020; Billieux et al., 2015) and whether platform affordances can cause such addictions (Roberts and David, 2025; Xavier and Tayana, 2016). Yet, the enforcement actions of the Commission against TikTok and Meta seem to assume a shared understanding of addiction and how platform design contributes to it.
In this blog post, we use these enforcement actions as case studies to understand how, in the absence of a scientific consensus on the causes of social media addiction, the Commission blurs policy discretion and empirical evidence—seemingly in application of the precautionary principle. To do so, we trace how different actors frame social media addiction and show how the Commission attempts to flatten competing accounts into a single, objectively verifiable fact.
The interface of science and policy in the DSA
The findings are ostensibly based on Articles 34 and 35 DSA, which require VLOPs to assess and mitigate systemic risks stemming from the design or functioning of their services, including risks to the physical and mental well-being of users. Notably, the operative provisions of the DSA do not mention “addiction”, meaning the Commission’s cases seemingly lean on Recital 83, which gestures at addiction without defining what that means or how it would be established. The findings are based on the platforms’ systemic risk reports, the interpretive guidance of the European Board for Digital Services (EBDS), and the Commission’s own autonomous investigative powers. The European Parliament sits outside the DSA’s enforcement framework but has been an active rhetorical force in producing resolutions that explicitly describe platform features in addiction-framed terms, often drawing on behavioural and neuroscientific language.
The DSA institutionalises a multi-layered information-sharing mechanism involving public and private actors who shape the evidence and understanding of systemic risks, whereby both scientific and policy input contribute to shaping the understanding of social media addiction in particular. Recital 90 is explicit in this matter as it requires VLOPs to consider “scientific insights” and consult with experts. Scientific expertise can also be directly consulted by the Commission or through the European Centre for Algorithmic Transparency.
That being said, the issue of addictive design requires policy choices to be made. Because of the lack of clear scientific evidence, the Commission may rely on the precautionary principle to make qualitative policy choices in order to safeguard important interests despite evidentiary uncertainty (Asselt and Vos, 2006; European Commission, 2000). It may also choose to privilege certain strands of evidence over others where the scientific community itself is divided.
Between evidence-based enforcement and policy judgements
The problem is not that the Commission is acting under scientific uncertainty; the problem is the discursive move that accompanies it, whereby preliminary findings and public statements are framed not as precautionary judgements made despite uncertain science, but as findings grounded in settled scientific fact about addiction. The legal mechanism invoked is precautionary, but the political communication surrounding it is declarative and assertive.
The Commission describes a policy judgement in the vocabulary of a scientific consensus that does not exist. At the same time, a lack of scientific consensus should not always be a barrier for political decision-making. This is precisely the case the precautionary principle foresees. Determining the existence of a (lack of) scientific consensus is an inherently contested endeavour as science is reflexive and contradictory by design, continually revising its claims through debate, replication, and new evidence (Jasanoff, 1990). To this must be added that a perceived lack of scientific consensus has, in modern history, been used by industry—most notably the oil and tobacco industries (Wolf et al., 2025; Michaels, 2008)—as a discursive strategy to avoid regulatory intervention that might otherwise prevent the continuous extraction of profit from harmful products. In the case of platforms, that they commercially profit from repeated and continuous user engagement is quite clear (Wu, 2025). However, the idea that they would intentionally and strategically mobilise the lack of scientific consensus over the addictive nature of digital architectures to avoid a more stringent enforcement of the DSA remains an open question.
From scientific uncertainty to competing narratives
Because the Commission operates with no explicit framework for the definition of addiction, the preliminary findings identify specific design features and treat their presence as sufficient evidence without requiring a diagnostic or causal demonstration that these features contribute to addiction. The evidentiary bar is behavioural and design-based, not clinical. However, a more contextual reading of these findings points to the interpretive work produced—at times in opposing directions—by the Parliament, the EBDS, and the platforms themselves.
In late 2023, the European Parliament issued a resolution in which it outlined its understanding of addictive design (European Parliament, 2023). The Parliament argued that “technology companies use design and system functionalities to take advantage of users’ and consumers’ vulnerabilities” (para. A) and that compulsive use of social media displays “similar side effects to substance-related addictions” (para. D). It bases these claims on an EPRS study which builds upon extensive empirical findings that show the addictive potential of particular platforms in relation to their interface, functionalities, and user demographics (Lopez-Fernandez and Kuss, 2019). Consequently, the Parliament’s resolution concludes that recommender systems “potentially represent an important persuasive, addictive or behavioural design feature” (para. M) and is “alarmed” that “certain platforms” exploit them “for commercial interests” (para. 3). It goes on to call upon the Commission to “adopt the necessary guidelines” under the DSA (para. 4). Despite the Parliament’s advocacy for evidence-based guidelines, the Commission has chosen the path of quick, theatrical, and legally contestable preliminary enforcement—crucially failing to provide legal certainty in the field.
However, in the press release for the Meta findings, the non-compliance claim is based on the issue of engagement maximisation induced by personalised recommender systems. These claims do resonate with the EBDS’s framing of addiction in its recently published report. The report almost exclusively considers evidence provided by consulted CSOs and VLOP reports—seemingly giving very limited consideration to empirical and scientific findings. Most of the affordances the EBDS considered as potentially leading to addiction match the risks described by the Commission in the press releases: endless feeds, autoplay and autoreplay of videos, and other design features maximising screen time such as personalised recommender systems.
At the same time, the systemic risk reports of four of the major VLOPs—Meta, YouTube, X, and TikTok—are quite telling in this respect, particularly as to the way platforms discursively frame and problematise addiction.
X’s report is the most superficial account. It only mentions addiction in passing under risks to public health (p. 58). Similarly, in its reports for Facebook and Instagram—which are nearly identical—Meta does not treat addiction or problematic use as a risk category within its own taxonomy, merely mentioning “problematic use” in its consideration of mental well-being (p. 15). In its assessment of the risks linked to its recommender systems specifically—one of the main features raised by the Commission’s preliminary findings—it claims to “optimise” these systems relative to risks by providing “users with more personalised and engaging recommendations” (p. 28). While X’s brief consideration of addiction does acknowledge that this may result from platform design (p. 58), Meta’s reports can be characterised by a clear refusal to engage with the notion that the platform itself or its design may lead to compulsive behaviour—consistently framing the responsibility for healthy engagement as resting with the users.
Conversely, YouTube and TikTok do explicitly acknowledge risks related to “addiction” (YouTube, p. 131) or “online engagement” (TikTok, p. 59) in which they go to some length in arguing that their perceived absence of a scientific consensus effectively means the platforms retain full discretion as to the mitigation measures it should put in place. While YouTube explicitly recognises that risks of addiction may arise from the design of the platform (p. 133), TikTok states quite plainly that it “does not believe that use of the Platform is inherently harmful” and then goes on to list different types of content that may be harmful “if viewed repeatedly” (p. 60). For its recommender systems, it directs the reader back to its description of those systems in which it offers a simple acknowledgement that they “may have an impact on a user’s extended use”, sending the reader to the online engagement section again (p. 20). On top of the content-centric framing of problematic use, TikTok’s report creates a closed loop of non-disclosure seemingly designed to simulate transparency while avoiding any substantive discussion of the addictive potential of the platform’s architecture.
Conclusion
At present, there is limited information about the evidence relied on by the Commission in reaching its preliminary findings. An analysis of the available documentation, combined with the discursive framing of addiction articulated by other institutional actors and VLOPs, reveals an interesting pattern. While scientific evidence on addiction caused by social media is contested and reportedly inconclusive, all institutional actors seemingly frame it as an objective reality, without engaging with the fact that there is scientific uncertainty. On the contrary, some platforms emphasise the lack of clear scientific evidence, which is, however, not reflected in the report of the EBDS, notwithstanding the fact that it often refers to the findings and claims of VLOPs in general.
The decision of the Commission to privilege direct enforcement action in the face of uncertainty and contestation is vulnerable to criticism, but is also, at least in part, understandable. Despite the lack of consensus on addiction, there could be sufficient evidence on the effects of platform design on mental health to justify the Commission’s enforcement if one considers both the precautionary principle and the best interests of the child. On the one hand, the precautionary principle justifies regulatory action in the absence of clear scientific evidence. On the other hand, the best interests of the child principle requires prioritising children’s safety and holistic well-being above other considerations. Their combination can warrant enforcement action, even when evidence is not yet settled science. If these principles are the driving force of enforcement, they are mainly the result of qualitative policy choices, rather than of clear scientific evidence. This is not necessarily wrong per se, but it should be clearly and transparently represented. The institutional discourse around addiction rather conveys that addiction and its evidence are objectively verifiable and uncontested. If enforcement action explicitly recognised the lack of conclusive evidence and the importance of making informed policy decisions, it would also weaken the narrative, often articulated by VLOPs, that the causes of social media addiction cannot be proven, and that the main drivers of systemic risks to minors are external factors and malicious actors rather than the platforms’ own interfaces.
