Meta, TikTok, YouTube, and Snapchat extract attention from children until it kills them.
It is true that the platforms describe themselves as neutral hosting environments, in the narrow sense in which a casino is a neutral hosting environment for the slot machines on its floor. The trouble is that the recommendation engine is not a neutral mirror of user intent; it is a continuously tuned objective function optimizing for time-on-site, and the objective function does not care whether the time it extracts comes from a healthy adult or a vulnerable thirteen-year-old.
What the complaint calls addictive is, formally, a variable-ratio reinforcement schedule delivered at machine speed. Cory Doctorow has described this continuous, computer-mediated adjustment of the user experience as twiddling, and the platforms do not merely show you content; they adjust the sequence of rewards and social validations in rapid, continuous micro-adjustments, hiding the payout schedule in a shell game conducted at machine speeds. When the user is an adolescent whose prefrontal cortex has not yet developed the inhibitory architecture to resist a variable-ratio schedule, the extraction is not a bug. It is the design operating exactly as specified.
The business model requires the extraction of attention, and the most profitable attention is the most captive attention. A kid who deletes TikTok still has Snapchat. A kid who quits YouTube still has Instagram. The platforms have spent a decade systematically disabling the four forces that historically constrained extraction: competition, via network effects and acquisition; regulation, via liability shields and captured legislatures; self-help, via anti-circumvention statutes that criminalize third-party clients; and labor, via the classification of their engineers as exempt professionals rather than workers with moral agency. What remains is a captive population, and the most captive population available is the one that cannot vote, cannot afford to leave, and cannot easily articulate the mechanism of its own capture.
The Social Media Victims Law Center filed this suit in Delaware on behalf of four families whose children died between July 2024 and September 2025. The teenagers came from Texas, North Carolina, Minnesota, and Tennessee. Their deaths are spread across fourteen months, and that window matters because by the time the first of these kids died, the industry had already been sued repeatedly over the same claims. Matthew Bergman, the founding attorney, noted it is particularly salient that these children died long after the initial suits were filed. This timeline overlaps precisely with the period when Meta appealed the verdict that found its platforms harmed a young user and YouTube settled a similar addiction lawsuit.
The salience is not merely temporal; it is structural. The platforms possessed the internal research documenting the mechanism of harm, and they chose to optimize the mechanism anyway, because the friction of stopping it was more expensive than the friction of settling the occasional wrongful death suit.
The legal system is now being asked to do the work that the regulatory state refused to do, which is to place a price on the extraction of adolescent attention. But a price is not a constraint when the margin on extraction exceeds the price. The remedy is not better content moderation or more transparent terms of service; it is the structural dismantling of the variable-ratio engine and the legal protection of adversarial interoperability, so that a third party can build a feed that does not optimize for time-on-site. The objective function does not contain a variable for a human life; it contains only a variable for time-on-site. The machine is working exactly as designed; it is the children who are failing to meet the specification.