In a federal courtroom in Oakland this month, a lawyer for the state of California read out a survey Meta had commissioned of its own users. More than half of teenagers on Instagram said they had experienced something bad or harmful on the platform in the previous seven days. The company had removed the offending content 0.02 per cent of the time. The survey had gone to Mark Zuckerberg personally, sent by Arturo Béjar, then a safety engineer, in 2021. He told the court he never heard back.

That single exchange, buried for years inside the company and now surfacing through discovery in a lawsuit brought by California and twenty-eight other states, is the most useful thing to have emerged from the trial so far. Not because it proves neglect, which Meta disputes, but because it demonstrates something more specific: that a platform can hold precise internal knowledge of harm to children while presenting a public face of continuous improvement. The gap between what a company knows and what it discloses is usually invisible. Litigation is one of the few processes that forces it open.

That gap matters in Britain for reasons that have nothing to do with the outcome of a Californian jury's advisory verdict. The Online Safety Act, now well into its implementation, asks platforms to assess their own risks to children and report accordingly. Ofcom's Protection of Children Codes came into force in July last year, and a further round of codes covering AI-assisted moderation is due this autumn. The regulator has fined firms, opened investigations, and pushed measurable improvements in areas such as age verification. But the architecture beneath all of it still depends on platforms producing honest risk assessments of their own products. Ofcom reviews and challenges what it receives. It does not generate the underlying data itself.

Meta's trial suggests what that dependency can look like from the inside. A 2021 survey showing majority-level harm, with a removal rate close to zero, sat within the company for years before a lawsuit dragged it into public view. There is no reason to assume Ofcom currently possesses anything equivalent about any platform operating in the UK, because the entire premise of the model is that firms voluntarily surface this kind of finding before a regulator has to go looking for it. The Oakland trial is a natural experiment in what happens when that assumption fails.

For the NHS, the practical consequence is not speculative. The children's commissioner's most recent annual report, published in June, recorded that referrals to child and adolescent mental health services passed one million for the first time in 2024–25, a rise of close to ten per cent in a single year. More than a third of children referred were still waiting for treatment by the following spring, and over 60,000 had been waiting more than two years. Clinicians working in these services routinely cite social media use as a contributing factor in presentations of anxiety, self-harm and body image distress, even where it cannot be isolated as a sole cause. CAMHS absorbs whatever proportion of that harm eventually reaches clinical thresholds, regardless of whether the platforms involved disclosed the underlying risk honestly or at all.

This is not an argument that Britain should import American-style litigation as a substitute for regulation, nor that Ofcom has been asleep at its post. It has fined companies and forced changes that a purely self-reporting model would not have produced. The argument is narrower and more uncomfortable: a regulatory system built on voluntary disclosure has no reliable way of knowing what it is not being told, and the only mechanism currently proving capable of extracting that information is adversarial litigation on another continent. Until Ofcom's autumn codes, or some future enforcement action, demonstrate that compliance data can be independently verified rather than merely reviewed, the health service will keep treating the downstream cost of a risk picture it was never shown.