The Specific Event

The European Union has formally targeted TikTok's recommendation algorithm under the Digital Services Act, raising the prospect of fines reaching 6% of TikTok's global annual revenue. The stated concern is what regulators are calling the "rabbit hole" phenomenon: a pattern where the recommendation system progressively serves users content that deepens psychological engagement at the expense of wellbeing, with documented mental health consequences particularly for younger users. TikTok now faces a structural choice. It can treat this as a compliance problem, make surface-level adjustments, and absorb whatever fines result. Or it can treat it as what it actually is: an organizational problem about how algorithmic systems distribute accountability when outcomes are harmful but causation is diffuse.

Why "Rabbit Hole" Is the Wrong Frame

The rabbit hole metaphor implies that users fall into something. It locates the problem spatially, as if TikTok's algorithm is a passive terrain feature and users are simply unlucky travelers who stumble. This is topography thinking, and it systematically misleads regulators. The more precise frame is topological. What TikTok's recommendation engine actually does is shape the structure of constraints governing what content a user encounters next. Each recommendation is not a neutral suggestion - it is a constraint that narrows the subsequent possibility space. The "hole" is not a place. It is a dynamic narrowing of structural options that the user experiences as engagement but that the organization experiences as retention metrics. These are not the same thing, and conflating them is precisely the analytical error that lets platforms avoid accountability (Kellogg, Valentine, & Christin, 2020).

The Accountability Diffusion Mechanism

What makes this case theoretically interesting is how accountability becomes distributed across organizational layers in a way that makes attributing harm genuinely difficult. The engineers who trained the engagement model did not intend to create mental health risks. The product managers who set retention KPIs did not instruct the algorithm to serve distressing content. The executives who approved the platform architecture were responding to competitive pressure from comparable platforms. At each layer, the decision appears locally rational. The harm emerges from the interaction of these decisions, not from any single one. Rahman (2021) describes this as the "invisible cage" dynamic: workers and, in this case, users operate under structural constraints they can perceive but cannot clearly attribute. The EU is attempting to hold TikTok accountable for a harm that the organization can plausibly claim no single actor within it chose to produce.

What Algorithmic Literacy Research Predicts Here

Research on algorithmic literacy reveals a consistent and uncomfortable finding: awareness of algorithmic systems does not translate to improved outcomes for people operating within them (Gagrain, Naab, & Grub, 2024). Users who know TikTok's algorithm prioritizes watch time and completion rates do not, as a result, consume content differently in ways that protect their wellbeing. This is the awareness-capability gap applied at scale. Knowing the structural feature exists is not the same as being able to respond effectively to it. TikTok's regulatory defense will likely lean heavily on the premise that informed users can manage their own consumption - a premise that the empirical literature on algorithmic literacy directly contradicts.

The DSA as a Schema Intervention

The more productive way to interpret what the EU is attempting is as a forced schema intervention at the organizational level. The Digital Services Act does not ask TikTok to change individual recommendations. It asks TikTok to disclose the structural logic of how its recommendation system operates and to demonstrate that systemic risks are assessed and mitigated. This maps reasonably well onto what schema induction does in learning contexts: rather than specifying correct procedures, it targets the structural features that generate outcomes across cases (Gentner, 1983). The regulatory theory, whether or not the drafters would use this language, is that procedural compliance is insufficient. What is needed is organizational understanding of the structural properties of the system being governed.

What Comes Next and Why It Matters

If TikTok responds with procedural adjustments - content warnings, screen time nudges, opt-out toggles - without addressing the structural logic of the recommendation engine, it will satisfy the letter of near-term regulatory demands while leaving the underlying accountability problem intact. The more consequential outcome of the EU investigation is not the fine. It is whether regulators develop enough structural understanding of how recommendation systems distribute harm to write governance frameworks that cannot be satisfied by surface-level procedural changes. That is a genuinely hard organizational theory problem, and it is one that existing compliance frameworks are poorly equipped to handle (Hancock, Naaman, & Levy, 2020).

References

Gagrain, A., Naab, T., & Grub, J. (2024). Algorithmic media use and algorithm literacy. New Media & Society.

Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155-170.

Hancock, J. T., Naaman, M., & Levy, K. (2020). AI-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication, 25(1), 89-100.

Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, K. S. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.