The UK government have announced a ban on social media for under-16s, taking effect Spring 2027. The ban is a significant step. It is also, on its own, incomplete. It addresses who can access social media. It does not address what social media does to those who do, including the children who will find ways around the ban, and the adults the ban does not cover. The deeper problem is that engagement-driven systems remain profitable when they amplify the material most likely to grip users. A duty of care sits alongside a profit motive pulling the other way. Mihaela Popa-Wyatt argues that losing that gap will require more than rules; it will require a price.

• The ban targets access, not the system producing the harm. Systems will continue to amplify content optimised for engagement for every user the ban does not reach.
• The Online Safety Act addresses risk but not incentives. Even where platforms comply with the Act and the ban, the business model that profits from harmful amplification remains intact.
• A Digital Harm Levy would close the gap. A tax calibrated to a Harmful Content Exposure Rate would make safer platform design economically rational, complementing the ban and the Act rather than replacing them.

The harm is engineered

Recommender systems decide which feeds a message enters, how prominently it appears, who sees it, and whether to push it further if early engagement is strong. None of these choices are forced by the content. They are the work of systems whose function is engagement, written into code and tuned on past data.

This matters because engagement and harm are correlated. Material that provokes outrage, fear, contempt, or in-group hostility tends to hold attention. Systems optimised for attention therefore amplify some kinds of speech more than others—at scale, repeatedly, and in personalised ways that compound over time.

The evidence is now substantial. The Internet Watch Foundation has documented a 26,362% year-on-year increase in photorealistic AI-generated child sexual abuse material. Despite age restrictions, an estimated 72% of children aged 8 to 12 bypass safety controls to access adult-tier platforms, where recommender systems routinely surface content on self-harm and disordered eating. Civil verdicts in the United States, including the Meta proceedings in New Mexico and the K.G.M. trial in Los Angeles, have begun to treat platform design itself as the proximate cause of harm to minors.

What the Ban does, and does not, do

The under-16 ban is the right response to part of the problem. Children are particularly vulnerable to engagement-driven amplification, and the evidence supporting protective intervention is strong. But the ban faces three limitations the policy debate should engage with directly.

First, enforcement. The same 72% bypass rate that justifies the ban also predicts its partial failure. Children who want access will find it through VPNs, family accounts, and the well-documented permeability of age-assurance systems. The eSafety Commissioner’s first compliance report found that “roughly 70% of children who had accounts before the ban retained access to at least one platform three months later, with no discernible reduction in cyberbullying or image-based abuse complaints from under-16 users.” Bypass rates thus remain high under enforcement. A ban that does not reach its target audience cannot eliminate the harm; it can move it to spaces where it is less visible.

Second, scope. The ban applies to children. It does not apply to the adults and older teenagers whom engagement-driven systems also harm, e.g., through coordinated harassment, ideological radicalisation, addictive use, and informational distortion.

Third, and most fundamentally, the incentive gap. The ban changes who can access social media. It does not change what social media is incentivised to do. Platforms will continue to operate the same recommender systems for every user the ban does not reach. Even where they fully comply with the ban and the Online Safety Act, the financial incentive to amplify harmful content remains intact, right up to the legal line.

These are not arguments against the ban. They are arguments for what must accompany it.

A Digital Harm Levy

Research at The University of Manchester proposes a Digital Harm Levy: a progressive tax on platforms calibrated to the rate at which their systems deliver harmful content to users—the Harmful Content Exposure Rate (HCER).

The HCER is the proportion of total user impressions on a platform that fall within harm categories defined by Ofcom. It measures what recommender systems actually deliver, not whether any particular item should be removed. A platform that reduces its HCER pays less. A platform that increases it pays more. The tax targets what economists call negative externalities, that is, the social cost the platform’s systems impose on third parties.

The levy is designed to complement the under-16 ban and the Online Safety Act, not replace them. Where the ban addresses access and the Act addresses specified harm categories, the levy addresses the residual externality through a continuous price. The three instruments do work the others cannot.

Crucially, the levy does not require platforms to remove content. To reduce their HCER, platforms can downrank harmful material, slow the spread of borderline content, redesign recommender systems, or invest in better user controls. The apt metaphor is throttling, not censorship.

Policy actions – and implementation

• Legislate a Digital Harm Levy alongside the under-16 ban. Introduce a progressive tax on digital platforms based on their Harmful Content Exposure Rate, targeting the systemic amplification of lawful-but-harmful content for all users.
• Build the auditing infrastructure. Mandate Ofcom to define auditable harm categories and work with standards bodies like the British Standards Institution and the Financial Reporting Council, to establish a methodology by which independent auditors can calculate the HCER from anonymised platform data.
• Phase implementation. Roll out the levy in three stages: definition of audit standards, a twelve-month pilot during which platforms report HCER data without financial penalty, and full enforcement integrated into the existing Digital Services Tax.

The ban makes certain online harms unreachable for children. A Digital Harm Levy would make them unprofitable for platforms. Speech would remain permitted and amplification ceases to be costless.