AI, Platforms & Digital Harm

A young person harmed after sustained interaction with a product engineered to sustain the interaction. Exploitation and sextortion enabled by design choices about who is permitted to contact whom. Recommendation systems that escalated a child toward material the operator already knew they escalated people toward.

Also the physical end of the same problem: autonomous and semi-autonomous systems, driver assistance, robotics, and algorithmic decision systems that injure someone or deny them something that matters.

These cases involve families in the worst circumstances of their lives. They are worth handling carefully rather than loudly, and nothing about this category is served by publicity.

The distinction that decides most of these cases is between a claim about content a third party created and a claim about the product itself. The second is ordinary product and negligence law: what the system was designed to do, what defaults it shipped with, what age assurance it had, what safety testing preceded release, and what the operator learned afterwards and left unchanged.

Framed that way, the questions are familiar ones. Was a safer design feasible. Was the risk known internally before release. Was the warning adequate to the danger. Was the product tuned for a metric the operator understood to produce this outcome.

The law here is genuinely unsettled and moving quickly, and it would be dishonest to present it otherwise. Courts have divided on where the line falls, several of the significant decisions are recent, and a case brought today is brought into an area still being decided.

Almost nothing that proves one of these cases is held by the person harmed. Model versions and their evaluation history, safety and red-team testing, moderation queues and escalation records, experiment results, internal research into the effect of a design change, and the server-side logs of the interactions themselves all sit with the defendant.

All of it runs on retention schedules set for ordinary business reasons, and those schedules keep running whether or not a claim is coming. Preservation is the first step, and it is time-sensitive in a way that has nothing to do with the limitation period.

A deleted account is not deleted evidence. The device matters too: it should be preserved rather than wiped, reset or handed in for repair, and examined by someone who will not destroy what is on it.

This category walks into general practices constantly and is very rarely recognised as a case. A family arrives describing what happened to a child, and the file it most resembles is not anywhere in the office.

The workup needs technical discovery, preservation aimed at systems rather than at documents, and a theory built for a developing area of law. Category fourteen covers the same ground where a matter has no obvious heading at all.