The Feed Nobody Inspected

Under the Online Safety Act, content that encourages suicide or self-harm is in the most serious category of material that platforms must prevent children from encountering. The duty is not a guideline. Ofcom, the regulator, has published codes saying how it should be met. Last week in Norfolk, a fourteen-year-old girl died after viewing suicide-related content on TikTok. Her case is now associated in public reporting with the name Adriana Retyte, and it has produced calls for UK watchdogs to crack down. Before we join those calls, we should ask a narrower question. What did anyone in authority actually see of what she saw?

I suspect the honest answer is very little. That gap is my subject.

The usual argument runs that the law exists, so the failure must be one of enforcement, so enforcement should be harsher. Larger fines, faster deadlines, sterner letters. None of this is wrong. But it assumes the regulator already knows the conditions it is regulating. Much of what a regulator learns about a platform comes through the platform itself: its risk assessments, its transparency reports, its account of how its recommendation systems behave. That is rather like a factory inspector who never walks the floor and instead reads the owner’s description of the ventilation. The description may be sincere. The owner may believe it. It is still not an inspection.

I want to be careful here. The facts we have establish that a child viewed this content on TikTok. They do not establish why it reached her, whether she searched for it, or how much of it a recommendation system chose to show her. Two explanations are plausible and should be held side by side. One is that the platform’s systems amplified material its own policies prohibit. The other is that such material slipped through moderation in volumes no system catches cleanly, and reached her by a route its designers never anticipated. These explanations call for different remedies. A regulator that cannot tell them apart will choose between them by temperament rather than evidence.

The strongest objection to my argument deserves a full hearing. Content moderation at this scale, its defenders say, is a problem of numbers, not of will. Every hour brings more uploads than any team could watch. Coded language changes faster than any classifier can learn it. Direct observation by a regulator would be a teaspoon held against a flood. On this view, the sensible course is to demand good systems and audit their outputs statistically, which is roughly what the current regime does.

My answer is that this objection is about catching every harmful video, and nobody needs to catch every video in order to learn what a fourteen-year-old’s feed looks like. That is a much smaller question, and the method suits it. Accounts registered as children, operated by trained staff, used the way children actually use them, would show a regulator what the recommendation system does with a lingering glance at a sad video, a late-night session, a search that edges toward despair. Researchers and journalists have run versions of this exercise informally. Nothing prevents Ofcom from making it routine, unannounced, and published, so that the figure reported each quarter is a count of what inspectors observed on screen and not what the company asserted.

Testimony belongs inside that inspection too. The people who know these feeds best are teenagers, and they describe them in their own terms: what keeps reappearing, what they scroll past, what they cannot stop seeing once it has appeared. Parents in Norfolk and everywhere else have something to tell a regulator as well. They know the hour the phone goes quiet in the bedroom and the evening a child stops coming down for supper. A consultation that collects these accounts systematically would be worth more than another round of written submissions from the companies under scrutiny.

The practical demand, then, is specific. Ofcom should hold standing powers to operate child-profile test accounts on regulated services and should publish the findings by platform. Companies should be required to give inspectors the recommendation history behind any reported case. Families such as the one in Norfolk should have a defined route to put their account into the evidence. Fines can follow the findings. They should not take the place of finding things out.

Somewhere in a regulatory office there is a report describing TikTok’s safeguards for teenage users. Somewhere in Norfolk there was a phone, held close to a face in the dark. The inspection we need is the one that looks at the second and checks it against the first.

If this subject affects you, Samaritans can be reached free in the UK and Ireland on 116 123, at any hour.