The Algorithm Doesn't Know What You Need. It Knows What Keeps You Here.
Written by Lea Nierlich and Tatum Kennedy
Inside the trial that could decide whether Meta has to answer for how it built Instagram
On August 18, 2026, a trial began in Oakland, California, that could have ended with Meta paying up to 1.4 trillion dollars, or facing something the company may have feared more: a court order forcing it to strip out the very features built to keep people, including children, scrolling for as long as possible. Twenty-nine states argued that Meta knowingly designed Instagram and Facebook to be addictive to young users, and knew the harm that caused. It never reached a verdict. On August 26, eight days in, Meta agreed to settle: 17.1 billion dollars to the states over ten years, along with binding changes to how Instagram and Facebook work for anyone under 18.
The trial's first witness was Arturo Béjar, a former Meta engineer who spent eight years working on product safety. He testified that Meta tracked internal data on how often teenagers reported experiencing bullying, self-harm content, and violence on its platforms, and that instead of making those numbers public, the company published a different, softer metric. "I think that these numbers create a false impression of safety," he told the court. Meta reached its settlement before its own chief executive had to take the stand. It is a familiar pattern: a company's own research confirms a harm, while its public messaging denies it. The pattern traces back to late 2021, when another former Meta employee, Frances Haugen, leaked thousands of pages of the company's own internal research, the Facebook Files, on exactly this question. The company knew. In public, it dismissed the concerns anyway.
What the algorithm is actually doing
A social media feed is not a neutral mirror of what is happening in the world or in your life. It is a prediction engine. Every second spent on a post, every video rewatched, every scroll not stopped, feeds back into a system constantly learning one thing: what will keep you here a little longer. It does not know what you need. It knows what works on you, specifically, and adjusts in real time to keep delivering it.
That matters because of how the brain responds to it. Notifications, likes, and comments trigger small hits of reward, and because you never quite know when the next one is coming, the pull to check again is stronger than if the reward were predictable, the same principle that makes slot machines effective. A developing teenage brain, more sensitive to reward and less equipped to regulate impulses, finds that pull even harder to resist. Researchers now estimate that roughly a quarter of adolescents meet the clinical criteria for social media addiction, with heavier use linked to worsened impulse control, rising rates of body image disorders from constant exposure to filtered images, and, because the algorithm rewards whatever holds attention regardless of whether it is good for the viewer, more extreme and risky content pushed toward the users most vulnerable to it, children and adults alike.
A crisis that is already visible
None of this is theoretical. Youth mental health has been declining sharply across Europe for years, roughly in step with the rise of algorithm-driven platforms. That timing is a correlation, not proof of a single cause, and scientists still debate how much of the decline is attributable to social media specifically, a debate that has played out in courtrooms too, where companies have argued there is no definitive scientific proof of a causal link. But the scale of the overlap is hard to ignore. In Germany alone, well over a million young people are now considered at risk. In Austria, more than a third of young people say they have come across suicide-related content online. Across the continent, only a small fraction describe themselves as genuinely happy.
One of the starkest examples: when a young person searches for self-harm content, the same recommendation logic applies as to anything else, surfacing more of the same, which can normalize or glorify it. The algorithm has no sense of what is dangerous to amplify. It only knows what a person is likely to keep watching, and serves exactly that, whether it helps them or hurts them.
The same logic shows up beyond individual mental health, too. An algorithm rewarding whatever holds attention tends to reward outrage over nuance, because anger and conflict keep people watching more reliably than calm, accurate information does. A 2026 report from the European Commission's own Joint Research Centre found this measurably eroding public trust in institutions across the EU. This is not a side effect platforms discovered too late, either. The same Facebook Files Haugen disclosed in 2021 had already shown the company's own engagement-based ranking amplified anger and misinformation, years before regulators caught up.
Why this has been so hard to sue over
For years, the biggest obstacle to holding platforms accountable was not proving the harm. It was two legal protections built for a different purpose: one shielding platforms from liability for content their users post, and the other protecting free speech, which courts have often extended to a platform's own choices about what to recommend. Suing over what shows up in someone's feed looked, for a long time, like suing over speech itself, a fight platforms were built to win.
The states suing Meta took a different route. They sued over the design, not the content: autoplay, infinite scroll, and notification systems engineered to pull people back in. That reframed the case closer to a product safety claim than a speech claim, more like suing a car manufacturer over a faulty brake than a publisher over a story. It is a path that had already worked once, in an earlier, related case: a jury found Meta liable on this exact theory, after seeing an internal document stating plainly that to win big with teenagers, Meta needed to bring them in as tweens, younger than the platform's own minimum age of 13.
The Oakland settlement turned that theory into product requirements, of a specific kind. Users under 18 in participating states now face a default two-hour daily time limit across Instagram and Facebook combined, a nightly blackout from midnight to 6 a.m., and muted notifications during school hours, on by default, though parents can lift the limit. Roughly 12 billion dollars of the settlement is unconditional. The remaining 5 billion is structured as leverage over the rest of the industry: Meta only pays it, and matches it, if Snap, TikTok, and YouTube adopt similar limits of their own. None of it touches the recommendation system itself. The settlement limits how long a teenager can be on the app. It does not change what the app shows them while they're on it.
Béjar himself, asked about the deal he had just testified toward, called it a significant milestone, but not a reason for parents to consider Instagram suddenly safe. His concern: the agreement, he said, "allows Meta to define harm" on its own terms. Other critics point out that age verification, meant to actually trigger these limits, still struggles to catch teenagers who simply enter a false birthdate, and that the settlement applies only in the United States.
Europe is testing a version of the same idea through regulation instead of lawsuits. Under the EU's Digital Services Act, major platforms must now assess and reduce the risks their recommendation systems pose to minors, and mitigate broader systemic risks their design creates, with regulators empowered to demand real changes to how a platform works, not just that certain posts come down. The first visible consequences are already appearing, not as fines yet, but as bans: France has moved to restrict social media access for under-15s, similar limits are under discussion in Spain, and momentum is building at the EU level. It is a slower, more preventive tool than a courtroom settlement, but rests on the same premise Oakland just confirmed: how a platform is built is something governments can hold companies accountable for, not just what it lets people post.
Why this matters beyond one courtroom
What ties this together is simple: these platforms are built, deliberately, to hold attention by any means that work, and the evidence from trials like this one shows the companies behind them knew what that would cost, especially for children, and kept building it anyway. That is not an accident of technology. It is a design choice, made by people who could have chosen differently, that caused real, measurable harm.
This is exactly the kind of decision AllRise exists to challenge: one made deliberately, inside a powerful organization, that causes documented harm to people, often the most vulnerable among them, and that the legal system was never quite built to catch. It cannot simply be looked away from because the harm is diffuse, or the company large, or because no single law was written with this exact situation in mind. Meta did not have to admit it was wrong to reach this settlement, and it still hasn't. It still had to change the product, at least around the edges of it. Thousands of individual lawsuits are still working their way through courts, and the recommendation system at the center of all of this remains untouched. This is not the end of the fight for accountability here. It is an early, partial result in a much longer one. Finding, or building, the legal path to get there is the work.
Get involved. Whether through funding, legal expertise, or evidence, every contribution helps determine which cases reach a courtroom and which do not. Support our work or explore our legal infrastructure.
Disclaimer: This article was developed with the assistance of artificial intelligence (AI) tools, which were used to support research, fact-checking, and editorial review. All content has been reviewed and approved by the author(s), who remain responsible for the accuracy, interpretation, and conclusions presented.