On August 12, 2026, jury selection began in Oakland, with opening statements set for the 18th. Twelve people will spend the coming weeks deciding a case about something none of them has ever held. No post is on trial, no video, no comment beneath a photo. What is on trial is a movement of the thumb.
Four states are bringing this first trial; twenty-nine have sued in all. The figure in the room runs as high as 1.4 trillion dollars — about what the company Meata is worth on the market. The number explains little. It marks the largest conceivable claim, not the expected bill, and it will still headline every story for weeks.
The more interesting number is smaller. It is six million, and it comes from a case already decided.
* * *
On March 25, 2026, a Los Angeles jury awarded a twenty-year-old plaintiff that sum — liability split seventy percent to Meta, thirty to Google. Against the scale of the two companies, whose equity was put at 217 and 415 billion dollars in court, the amount is a rounding error. What mattered was how the verdict was reached.
The plaintiff’s lawyers had chosen to keep the platforms’ content largely out of the case. Almost nothing about what appeared on Instagram, who had posted it, what the algorithm pushed to the top. What was argued instead: infinite scroll, notifications timed to moments of flagging use, counters beneath images that make visible how many others have already approved. What was argued were decisions made in design reviews.
Legally, that is a different category. In the United States, platforms enjoy broad protection for third-party content, and where content is in play, free speech can be raised on top of it. A feature is not speech. A feature is a product, and products come with negligence, duties to warn, and product liability — a vocabulary long assumed to belong to car seats and medicines.
* * *
The defense’s strongest argument, meanwhile, lies not in law but in psychiatry. The diagnostic manual treated worldwide as the reference recognizes no such thing as social-media addiction in its current edition. Anyone asked to answer for inducing a condition may fairly ask which condition is meant.
Behind that sits a dispute that has run for years in the research and remains unsettled. On one side, work finding associations between heavy use and depressive symptoms in adolescents, girls especially. On the other, researchers such as Candice Odgers, Amy Orben, Andrew Przybylski, and Christopher Ferguson, who take the same datasets apart and find effect sizes that barely register beside sleep loss or family conflict. Ferguson’s meta-analysis concludes that the experimental evidence does not support causal claims. Rausch and Haidt reply that he blended methods, durations, and outcomes until the result became unreadable. Choosing the side that suits your prior does not resolve the dispute. It only tunes it out.
* * *
Whether a mechanic makes people ill over the long run is contested. Whether it works in the short run is not.
The UK’s financial regulator placed some 9,000 people in a simulated trading environment and varied individual design elements. Push notifications raised the number of trades by eleven percent; a points-and-prize scheme raised it by twelve. Participants between 18 and 34 shifted their portfolios toward risk more than older ones did under nearly every element tested. These are not soft correlations drawn from self-report, but a randomized experiment with a sample unusually large for questions like these.
One finding from the same study weighs more than the percentages. The elements kept people in the app longer and moved them to act more often — yet did not lead them to open the underlying information on fees and risk any more frequently. The added attention flowed into the motion, not into understanding what the motion was about.
That is the actual point, and it has little to do with malice. Effect known, consequence unknown: a team can say with precision that an element lifts the completion rate, and with little precision what that element set off in the person it worked on. The first can be read off a dashboard. For the second, there is usually no column.
Turning off an element that works rarely earns anyone credit.
* * *
What is being argued in Oakland concerns a handful of platforms. The mechanic behind it does not. There is no need to overstretch the analogy — a trading case and a youth-protection case are different things — but the design repertoire is the same. The progress bar that makes a half-finished profile unbearable. The push message that arrives when use tapers off, not when something important happens. The red ticker that pushes you to act where waiting would often be the better call. The confetti after the transaction. The list showing what others are buying right now.
The same elements that make an app hard to put down make an investing app hard to sit still with. Only the stakes differ. In one, time is on the line; in the other, savings. And in finance the vocabulary shifts as well: what a social app calls “engagement” is eventually called, before a regulator, “conduct that was not in the customer’s interest.”
Regulation is already moving. The EU’s Digital Services Act already prohibits manipulative interface design by large platforms. The Digital Fairness Act, announced for late 2026, names “addictive design” for the first time — meaning, explicitly, infinite scroll, autoplay, and notification patterns. In Switzerland, a draft platform law is out for consultation that child-protection groups consider too weak. The pace and the reach are open to argument. The direction is not.
* * *
Here the obvious objection is due, and it deserves to be taken seriously: measuring effect is exactly what these firms do. They test, optimize, iterate. They are not short on measurement — they measure constantly. True, and it only moves the point more precisely to where it belongs. What gets measured is one goal: completion, time spent, return visits. What is missing is not the measurement but the second goal to hold the first against — a figure that captures whether the action triggered served the person or only the metric. Without that second yardstick a team is not clueless. It is informed on one side, with high precision.
That one-sidedness is what turns expensive in court. Not the bad intent, which is hard to prove, but the documented diligence in exactly one direction. Every A/B protocol showing how well an element lifted the number also shows that someone was watching closely — just not the consequences.
What is left is a plain shift, and it reaches everyone who designs surfaces, not only the defendants in Oakland. For a long time the rule was: you are responsible for what you say. Now a second sentence is forming beside it. You are also responsible for what you build.
* * *
This essay is part of a longer inquiry into how fluent systems and frictionless interfaces produce actions no one examines any more — and what that means for the people who build them.
If that stays with you, I write about it now and then: Newsletter.
And if you work in financial services, I’ve worked the same diagnosis out for that industry in a German-language whitepaper, «Journey Economics — Wo das Geld wirklich liegt»:
