On a trial began in Oakland that was never argued to the end. Four states stood in for 29 against Meta, before a federal judge and eight jurors whose verdict was to be advisory only. Eight days later, with Instagram chief Adam Mosseri due to take the stand for a second day, a settlement was on the table. The judge approved it the same day.
The headlines belonged to the number. Meta will pay around $12 billion over ten years, up to just over $17 billion if Snap, TikTok and YouTube take on comparable obligations and payments. The four suing states had demanded up to $1.4 trillion, a figure experts considered barely realistic. It carried the coverage of the first days of the trial and explained nothing.
More interesting than the number is a list.
Anyone under 18 will in future get a default daily limit of two hours on Instagram and Facebook that only parents can lift. Between midnight and six in the morning the app is blocked. During school hours notifications stay silent. Minors will no longer see how many likes are under a picture, and filters that simulate cosmetic surgery will disappear. An independent monitor checks implementation.
You can read this list as a child-protection programme. You can also read it as what it formally is: a catalogue of design decisions. Default settings, notification logic, counters, filters. Hardly any of it concerns what anyone posted. Almost all of it would normally be decided in a design review. Now it is in a consent judgment.
Meta admitted no wrongdoing, and the judgment states explicitly that it sets no standard of care and creates no precedent. Legally, that is correct; in practice, though, since 26 August there has been a document naming the design elements to be restricted for minors, backed by 51 attorneys general, well beyond the 29 states that sued. Anyone building a comparable element from now on will find it hard to claim they couldn’t have known.
That it came to this has to do with a verdict handed down in March that, measured in dollars, was almost nothing.
On a jury in Los Angeles awarded a twenty-year-old plaintiff six million dollars, half in compensatory and half in punitive damages. Liability is split seventy per cent to Meta and thirty to Google. Measured against the two companies’ equity, put at 217 and 415 billion dollars during the trial, the sum is a rounding error. Both companies have announced appeals. The route there was remarkable.
The plaintiff’s lawyers had decided to keep the platforms’ content largely out of the case. Barely a word about what could be seen on Instagram, who had uploaded it, what the algorithm pushed to the top. What was argued was infinite scroll, notifications timed for moments of flagging use, counters under pictures showing how many others had already approved. What was argued were decisions made in design reviews.
Legally, that is a different category. For third-party content, platforms in the US enjoy wide protection, and where content is involved, free speech can be invoked as well. The plaintiff’s line amounted to this: a feature is not an opinion but a product, and products are subject to negligence, duties to warn and product liability – a vocabulary that long seemed reserved for car seats and medicines. Whether this line will hold is disputed in the US. The platforms argue that curating a feed is protected expression too.
Five months later, exactly these features were the subject of a settlement worth tens of billions. The counters under the pictures, still evidence in March, are to be switched off for minors.
The defence took its strongest argument from psychiatry. The diagnostic manual that serves as the reference worldwide does not, in its current edition, recognise social media addiction. Meta maintains that the mental health of young people is too complex to attribute to a single app. Anyone who is to be held liable for causing a condition may, with some justification, ask which condition exactly is meant.
Behind this lies a dispute that has been running in research for years and is not settled. On one side, studies finding links between heavy use and depressive symptoms in adolescents, especially girls. On the other, researchers such as Candice Odgers, Amy Orben, Andrew Przybylski and Christopher Ferguson, who take apart the same datasets and find effect sizes that barely register next to lack of sleep or family conflict. Ferguson’s meta-analysis concludes that the experimental evidence does not support causal claims. Rausch and Haidt counter that he mixed methods, time frames and outcome measures in a way that makes the result unreadable. Anyone who picks the side that suits their own position has simply not listened to the dispute.
The Oakland settlement sidesteps this dispute. What was settled in the end was not whether Instagram makes people ill, but which default settings a product for fourteen-year-olds may have.
Whether a mechanism makes people ill is disputed. That it shifts behaviour, the UK Financial Conduct Authority has measured.
In 2024 it put just over 9,000 people into a simulated trading environment and varied individual design elements. Push notifications raised the number of trades by eleven per cent, a points-based reward scheme with a prize draw by twelve. Participants aged 18 to 34 shifted their portfolios more towards risk than older ones under almost all the elements tested. Unlike most findings in this debate, this rests on a randomised experiment with a sample unusually large for such questions.
One finding of the same study weighs more than the percentages. All four elements drew attention into the app; two of them led people to trade more. Yet they opened the information on past performance and risk no more often than the control group did.
Knowledge of effects without knowledge of consequences: a team can say with great precision that an element lifts the conversion rate, and with little precision what that element did to the person on whom it worked. The first can be read off the dashboard. For the second there is usually no column.
Switching off an element that works rarely earns praise. In Oakland it was agreed by contract.
What was settled in Oakland concerns a handful of platforms. The mechanics behind it can be found in many apps. There’s no need to overstretch the analogy – a trading experiment and a child-protection case are different things – but the design repertoire is the same. It ranges from the progress bar that makes a half-completed profile unbearable and the push notification that arrives when usage flags, to the red share price that urges you to act where waiting would often be the better investment decision. Then there is the confetti after completion and the list showing what others are buying right now.
The same elements that make a social app sticky make an investment app pushy. The stake differs. With one, time and well-being are at risk; with the other, savings. And in financial services the vocabulary shifts further: what a social app calls “engagement” will at some point be called, in front of a regulator, “behaviour that was not in the customer’s interest”.
Europe is further along than the debate around Oakland suggests. On the European Commission preliminarily found that TikTok’s design breaches the Digital Services Act. For the first time, such a finding targeted the architecture of a platform rather than content or data use. The Commission suggests gradually switching off infinite scroll, introducing effective screen-time breaks, including at night, and adapting the recommender system. Similar proceedings against Meta have been running since 2024.
The Digital Fairness Act, which is to address “addictive design” explicitly for the first time, has been announced for the fourth quarter of 2026 and has not yet been tabled. Unlike the Digital Services Act, it targets not only the largest platforms but business practices towards consumers in general. That brings it closer to the apps people use to invest their money.
In Switzerland, the consultation on the draft Federal Act on Communication Platforms and Search Engines closed in February. The draft mainly requires the largest platforms to be transparent about moderation, advertising and recommender systems; design that prolongs use is not at its centre. The Greens consider it clearly inadequate; consumer protection groups support its direction but want more.
One could object that these companies measure effects all the time. But the measurement serves one goal: completion, time in the app, return visits. What is missing is not the measurement but a second goal against which the first measurement would be held – a metric that captures whether the action triggered served the person or only the KPI. Without this second yardstick, a team is informed one-sidedly, with great precision.
Precisely this one-sidedness becomes expensive in court, because it is documented. In Los Angeles there was the added fact that the jury found malice, oppression or fraud for the punitive damages. In Oakland, Meta accused the states of cherry-picking from internal documents. The accusation shows where the evidence now lies: in what a company has written down about its own products. Every A/B test log proving how well an element lifted the number also proves that someone was looking closely – just not at the consequences.
The shift affects everyone who designs interfaces, not just Meta. For a long time the rule was: you are responsible for what you say. Since this year, a verdict and a settlement support a second sentence: you are also responsible for what you build.
This essay is part of a longer engagement with how fluent systems and frictionless interfaces lead to actions nobody checks any more – and what that means for those who build them.
If this resonates, I write about it at irregular intervals: newsletter.
Read more:
