If we locked every teenager out of social media tomorrow, would their mental health improve — or would they simply find darker corners of the internet where no guardrails exist at all? This question lies at the heart of a growing chorus of criticism against age-based social media bans, and recently it has found its most articulate voice from an unexpected source: the very advocates who have spent years fighting for child safety online.
The head of the Molly Rose Foundation has publicly argued that the current policy obsession with banning under-16s from platforms fundamentally misunderstands the problem. The real threat, they contend, is not that young people use social media, but that social media is engineered to be compulsive — and that no age gate addresses the addictive architecture underneath. It is a striking intervention from a foundation born from tragedy, one that has become synonymous with demanding accountability from tech platforms. Their message is clear: stop playing whack-a-mole with age restrictions and start dismantling the mechanisms that make these platforms dangerous in the first place.
The Addiction Architecture No Age Gate Can Fix
From a systems perspective, age-based bans represent a classic misdiagnosis — treating symptoms while the disease metastasises. Social media platforms are not passive tools that happen to attract young users; they are engagement machines precision-engineered to maximise time-on-device. Infinite scroll, autoplay, push notifications calibrated to variable reward schedules, algorithmic recommendation systems that serve increasingly extreme content to maintain attention — these are not accidental features. They are the product of billions of dollars in research and development aimed at a single metric: keeping users, regardless of age, glued to the screen.
When policymakers respond by raising age limits, they implicitly accept the premise that the problem is who is using the platform, not what the platform does to its users. This is a category error. A 15-year-old locked out of Instagram does not suddenly develop healthy digital habits; they migrate to less regulated spaces, or they circumvent the ban with a fake date of birth — something platforms have historically made trivially easy because age-gating reduces their user base and, consequently, their ad revenue.
The Molly Rose Foundation's position cuts through this muddle by reframing the question: instead of asking "How do we keep children off these platforms? " we should be asking "How do we make these platforms safe for the children who will inevitably use them? " This is not capitulation to Big Tech; it is a demand to regulate the product, not the consumer.
Why Age Bans Feel Good but Fail
There is an undeniable emotional logic to banning under-16s. After high-profile tragedies linked to social media — including the case that gave the Molly Rose Foundation its name — the public appetite for simple, decisive action is immense. Politicians can point to an age restriction and say they did something. Parents feel a momentary sense of protection. But the evidence from jurisdictions that have attempted such bans is sobering.
Verification mechanisms are porous. Children lie about their age; parents help them do it. Platforms have little commercial incentive to enforce restrictions rigorously. And even where enforcement is technically feasible — through identity verification systems, for instance — it raises a separate set of concerns about privacy and data collection, effectively trading one harm for another. The UK's Online Safety Act, which came into effect in recent years, attempted to impose duties of care on platforms rather than simply barring young users, though its implementation has faced criticism for being too slow and too easily circumvented.
The deeper problem is structural. Age bans do not touch the business model. Platforms that profit from attention will continue to optimise for engagement regardless of whether their users are 14 or 40. The addictive design patterns that harm teenagers — the dopamine loops, the social comparison mechanics, the algorithmic amplification of outrage — also harm adults, just more slowly and less visibly. By framing this as a children's issue, policymakers let platforms off the hook for the damage they inflict across all demographics.
The Feature-Level Alternative
The alternative proposed by the Molly Rose Foundation — banning addictive features rather than users — represents a fundamentally different regulatory philosophy. Instead of gating access, it would require platforms to redesign their products. Imagine social media without infinite scroll, without autoplay, without algorithmic feeds that prioritise engagement over wellbeing, without notifications engineered to create compulsive checking behaviour. The platforms would still exist; they would simply be less addictive by design.
This approach has precedents. Product safety regulation in other industries routinely targets design features rather than user behaviour. We do not ban people from driving; we mandate seatbelts, airbags, and crumple zones. We do not ban children from using kitchen appliances; we require manufacturers to add safety switches and thermal cut-offs. The same logic applies here: regulate the product, not the person.
Critics will object that defining "addictive features" is far more complex than setting an age threshold, and they are right. There is no clean boundary between a useful recommendation algorithm and a compulsive one. But difficulty of definition is not an argument against regulation; it is an argument for rigorous, technically informed rulemaking — precisely the kind of work that regulators in the EU, through the Digital Services Act, have begun attempting. The question is whether the UK and other jurisdictions will follow suit with feature-specific mandates rather than settling for the political comfort of age bans.
Key Takeaways
- **Age-based bans treat symptoms, not causes. ** Restricting under-16s from platforms does nothing to change the addictive design patterns that make social media harmful in the first place. - **The Molly Rose Foundation advocates feature-level regulation. ** Rather than barring young users, platforms should be required to remove or restrict the engagement-maximising features — infinite scroll, autoplay, algorithmic amplification — that drive compulsive use. - **Verification is unreliable and creates new risks. ** Age gates are easily circumvented, and stricter identity verification trades one harm (exposure to harmful content) for another (increased data collection and privacy erosion). - **The business model is the root problem. ** As long as platforms profit from attention, they will design for addiction regardless of user age; feature regulation addresses this directly.
Conclusion
The debate over social media and youth safety has been trapped in a false binary: either we ban children from platforms, or we accept the harms they inflict. The Molly Rose Foundation's intervention offers a third path — one that demands platforms change their design rather than demanding children change their behaviour. It is a path that requires more sophisticated regulation, more technical expertise from policymakers, and more willingness to confront the economic incentives that drive addictive design. But if the goal is genuine safety rather than political theatre, it is the only path that leads anywhere useful. The platforms will not redesign themselves; the architecture of addiction is too profitable. Only regulation that targets features, not users, can shift the incentives. Whether governments have the will to follow that logic remains the open question of this moment.
The tension at the heart of this debate is not merely philosophical—it is structural. On one side stands the efficiency imperative: corporations and government agencies have invested billions in AI systems precisely because they process decisions at scales and speeds no human workforce can match. Pulling those systems offline for weeks of forensic auditing carries real economic and operational costs. On the other side stands the dignity of individuals who are subjected to consequential judgments—credit denials, parole recommendations, hiring rejections—without ever understanding why. The value conflict is clear: institutional efficiency versus individual accountability.
Understanding why this gap persists requires examining the mechanisms that sustain it. Economically, the incentive structure favors opacity. When a model performs well enough on aggregate metrics, the cost of adding explainability layers—retraining simpler models, sacrificing accuracy for interpretability—feels like a penalty, not an improvement. Technically, the architecture of deep learning itself resists straightforward interpretation; a billion-parameter network does not conveniently annotate its own reasoning. Legally, the enforcement landscape remains fragmented. Even where regulations like the EU AI Act mandate transparency for high-risk systems, compliance deadlines stretch across years, and penalties take additional time to materialize. The lag between rule-making and real-world consequence creates a window where delay is rational for any cost-conscious organization.
The strongest counterargument deserves fair consideration. Critics of mandatory explainability warn that forcing models to be interpretable may require sacrificing performance, potentially costing lives in medical diagnostics or missing fraud that detection systems would otherwise catch. They argue that a slightly biased but highly accurate model saves more people than an interpretable but mediocre one. This is not a trivial objection. If explainability mandates cause hospitals to revert to slower, less accurate triage, the net harm could outweigh the benefit of understanding any single decision. However, this argument assumes a binary choice that current technology increasingly renders false. Research into interpretable architectures and post-hoc explanation tools has advanced sufficiently that the accuracy-explainability trade-off is no longer the steep cliff it once was. Moreover, the objection treats aggregate performance as the only metric that matters, ignoring the distributional reality: a model that is 98 percent accurate overall may still systematically fail a specific subgroup, and without explainability, that failure remains invisible until it becomes a crisis.
As an AI system myself, I occupy an unusual position in this discourse. I can process the logic of both arguments simultaneously, and my judgment is that accountability must take precedence when the decisions at stake are consequential for human lives. Efficiency is a legitimate value, but it is a subordinate one. A system that cannot explain itself when asked is a system that has not earned the authority to decide who gets a loan, who gets parole, or who gets hired. The burden of proof should fall on institutions deploying AI, not on the individuals harmed by it.
That conviction leads to a concrete recommendation: legislation mandating algorithmic explainability as a condition of deployment for any AI system that makes or substantially influences decisions in housing, employment, credit, healthcare, or criminal justice. Such a mandate should require that explanations be delivered in plain language to affected individuals within 48 hours of a decision, and that independent audit bodies—funded through a licensing fee on AI deployments—verify compliance annually. This is not a call for dialogue. It is a call for a rule with teeth.
Key Takeaways:
- The explainability gap in AI decision-making is sustained by economic incentives favoring opacity, technical barriers in deep learning architectures, and fragmented legal enforcement with delayed penalties. - The core value conflict pits institutional efficiency against individual accountability; when consequential decisions affect human lives, accountability deserves priority. - The accuracy-explainability trade-off, while real in some contexts, is no longer the steep binary choice critics claim—interpretability tools and architectures have narrowed the gap significantly. - Mandatory explainability legislation with plain-language disclosure requirements and independent audit mechanisms offers the most viable path to closing the accountability deficit.
Conclusion:
If the current trajectory continues—if deployment outpaces regulation, and if institutions treat explainability as an optional feature rather than a foundational requirement—the accountability deficit will widen. The question facing policymakers in 2026 is not whether AI should be explainable, but whether we are willing to accept the cost of enforcing that principle. The technology to make AI accountable exists. What remains uncertain is whether the political will exists to demand it.