September 3, 2026

Eighteen Billion Dollars, a Two-Hour Timer, and What We Actually Know About Kids and the Feed

Meta settled the biggest case ever brought by the states the morning after its Instagram chief testified, and admitted nothing. The money is a rounding error. The timer, the age machine, and a decade of evidence that cuts both ways are the story.

The facts:

• On August 26, in the second week of a federal trial in Oakland and a day after Instagram head Adam Mosseri testified, Meta settled with the states that accused it of designing Instagram and Facebook to hook kids; the agreement says it "does not constitute an admission by Meta of any liability, wrongdoing, or violation" (ABC News, settlement agreement)
• The deal is worth up to $17.1 billion over ten years: at least $12.1 billion guaranteed and $5 billion more only if YouTube and TikTok agree to comparable terms; a separate Texas settlement of about $1 billion brings the total near $18 billion; Meta's 2025 revenue was $201 billion (CBS News, Washington Examiner)
• The states had asked for $200 billion; Meta's own July filing said the demands "would amount to more than $1.4 trillion"; its share price rose in the hours after the deal (The Guardian, Ars Technica)
• For users under 18: a default two-hour daily limit across both apps that only a parent can lift, no access from midnight to 6 a.m., notifications muted 8 a.m. to 3 p.m. on school days, pauses at 15, 60 and 90 minutes, like counts hidden, cosmetic filters off, and an optional feed with no recommendation algorithm; messaging and video of 22 minutes or more don't count toward the two hours (Wired, CBC, settlement agreement)
• Most obligations take effect six months after the court's judgment; if Snap, TikTok and YouTube match, the limit drops to one hour per app and the night lockout expands to 10 p.m. to 7 a.m. (settlement agreement, Meta newsroom)
• Meta must build a system, audited by outsiders, that wrongly passes no more than 10% of 16 and 17 year olds and 3% of 13 to 15 year olds as adults; its current estimation reads posts, likes, log-in times and photos, including "height or bone structure" (The Verge, Fortune)
• Buried in the agreement: the states promised "fully, finally, and forever" not to sue Meta under children's privacy law over its use of kids' data to train that age model; the FTC, which enforces that law, is not a party (TechCrunch)
• Weeks earlier a New Mexico judge found Meta's platforms a public nuisance and ordered a $567 million fund on top of a $375 million jury penalty, $942 million in all; Meta is appealing; in March a Los Angeles jury found Meta and YouTube negligent and awarded $6 million to one 20-year-old plaintiff (Washington Examiner, CNBC)
• A 2019 internal Meta slide, published by the Wall Street Journal in 2021, read "We make body image issues worse for one in three teen girls"; Meta replied that on 11 of 12 well-being issues, teen girls who struggled said Instagram made things better rather than worse, and that the research "did not measure causal relationships" (Wall Street Journal, Meta newsroom)
• The US Surgeon General's 2023 advisory cited a study of 6,595 adolescents in which those on social media more than three hours a day faced double the risk of depression and anxiety symptoms, and concluded "we do not yet have enough evidence to determine if social media is sufficiently safe for children and adolescents" (HHS advisory)
• CDC surveys: the share of high schoolers reporting persistent sadness or hopelessness rose from 30% in 2013 to 42% in 2021, then 40% in 2023; among girls it was 57% in 2021 and 53% in 2023; the suicide rate for ages 10 to 24 rose 62% between 2007 and 2021 (CDC Youth Risk Behavior Survey, National Center for Health Statistics)
• Psychologist Candice Odgers, reviewing Jonathan Haidt's book in Nature, wrote that the claim of an epidemic caused by phones "is not supported by science," that most data are correlative, and that "young people who already have mental-health problems use such platforms more often" (Nature)
• Phone-log data cited in Nature this week put median Instagram use at about 13 minutes a day and all social media at about 80 minutes, which led one researcher to question a two-hour allowance (Nature)
• Australia's under-16 ban has been in force since December; a BMJ study found over 85% of under-16s still using the covered platforms months later; the UK announced its own under-16 ban for spring 2027, and the EU is pressing Meta to extend the US protections to Europe (BMJ, GOV.UK, France 24)
• A study of nearly 40,000 Ontario students in JAMA Pediatrics found 21% use AI chatbots for emotional support and those students were almost twice as likely to show clinical distress; the lead author says that is a link, not a cause; nothing in the settlement covers chatbots (CBC, New York Post)

The settlement got signed the morning Adam Mosseri was due back on the stand. That's the detail to hold onto. Meta had gone to trial twice this year and lost both times, in New Mexico and in Los Angeles. This was the federal one, the one with twenty-nine states and a filing that put its own exposure above a trillion dollars. It listened to its Instagram chief for a day, and then it paid.

Eighteen billion sounds like a lot until you set it next to two hundred and one billion in revenue last year. Meta is booking about ten billion of it as a charge this quarter, and the stock went up. A price the company can plan around, paid over ten years, with almost a third of it contingent on what TikTok and YouTube decide to do.

what Meta knew, and what it said back

The case rested on the company's own paper. In 2021 the Wall Street Journal published a slide from Meta's internal research, dated 2019, that said, in the company's words, "We make body image issues worse for one in three teen girls." Another slide said teens blamed Instagram for rising anxiety and depression, and that the reaction was "unprompted and consistent across all groups."

Meta's answer at the time is worth reading too, because it's the argument it made in court five years later. On eleven of twelve well-being issues, it said, the girls who struggled reported that Instagram made things better, not worse. And the research measured no causation at all. It was a survey of what teens believed about the app, not proof of what the app did.

Both of those things are in the record. Both are true. That's the shape of this whole subject.

the evidence, both ways

Here's the case that something happened. The share of high schoolers telling the CDC they felt persistently sad or hopeless went from thirty percent in 2013 to forty-two percent in 2021. For girls it hit fifty-seven percent. The suicide rate for ten to twenty-four year olds rose sixty-two percent between 2007 and 2021, after a decade of holding flat. The Surgeon General's advisory pointed at a study of six and a half thousand kids in which more than three hours a day on social media doubled the risk of depression and anxiety symptoms, and then said the thing agencies rarely say: we don't have enough evidence to call this safe.

Here's the case that the phone is the wrong suspect. Candice Odgers, who has spent a career on exactly this question, reviewed the book that launched the panic and wrote that the epidemic claim "is not supported by science." Hundreds of researchers have looked for the big effect, she said, and found a mix of nothing, small, and mixed. A seventy-two-country analysis found no consistent link between the rollout of social media and well-being. And the correlations that do show up run the other direction as often as not: kids who are already hurting use the platforms more. Andrew Przybylski, who ran the numbers on three hundred and fifty thousand adolescents, put the average correlation between screen time and well-being at about the size of the one between wearing glasses and well-being. A rounding error.

Haidt's reply is that the average hides the group that matters. Screens in general, all kids, tiny effect. Social media specifically, for girls, the correlation roughly doubles. That's not nothing. It's also not the epidemic the courtroom language implies.

Five hundred years ago, within a generation of the printing press, scholars were writing that the flood of books was ruining memory and judgment. They were wrong about the verdict and right about the cost. Reading did change what a mind did all day. The question with this one isn't whether the panic is a panic. It's what it's right about.

the two-hour question

So what does two hours actually buy? Less than the headline. Phone-log data cited in Nature this week put the median teen's Instagram use at about thirteen minutes a day, and all social media combined at about eighty. A cap at a hundred and twenty minutes, with messaging and long video exempt, doesn't touch the median kid. It touches the tail, which is the kid the case was about.

The parts of the deal with real evidence behind them are the quiet ones. Harvard's Christian Sandvig singled out the lockout from midnight to six as "really well-supported." Sleep is the one mechanism nearly everyone in this fight agrees on. Hidden like counts and the end of the cosmetic filters go at the comparison machine directly. An Australian researcher looked at the same list and said every item was about engagement, not safety. She's half right. Engagement is where the harm was alleged to live.

the machine that decides who's a child

None of it works unless the app knows who's fifteen. That's the part of the settlement with the most pages and the least attention. Within a year Meta has to run an age system, checked by outside auditors, that wrongly waves through no more than ten percent of sixteen and seventeen year olds and three percent of the younger ones. Its current tools read what you post, when you log in, who your friends are, and, in its own words, "height or bone structure" in your photos. Skip the check for two weeks and you're treated as a teen whether you like it or not.

To train that, Meta gets something in return. The states agreed "fully, finally, and forever" not to sue under children's privacy law over how it keeps and uses kids' data for the age model. The federal agency that actually enforces that law isn't a party to the deal. The Electronic Frontier Foundation's read is that the settlement embeds a flawed technology in everyone's account for ten years. Meta's read is that the app stores should be doing this instead. Apple and Google say no.

going forward

The clock is six months from the judge's signature for most of it. Then it depends on three other companies. If Snap, TikTok and YouTube match the terms, the limit drops to an hour per app and the night lockout grows, and Meta pays the other five billion. Meta's chief lawyer said it plainly: teens move across dozens of apps, so it only works industry-wide. Which is another way of saying a two-hour wall on Instagram sends a kid to the app without the wall.

The rest of the world already ran the experiment. Australia banned under-sixteens outright in December. By spring, a study in the BMJ found more than eighty-five percent of them still on the covered platforms, and Meta reported it had removed three quarters of a million accounts anyway. Britain announced its own ban for next spring. Brussels wants the US settlement extended to Europe and is holding a six-percent-of-revenue fine behind its back. Thirty-nine states now keep phones out of school, and the research there is as mixed as everything else: Norwegian girls visited the doctor less for psychological symptoms after bans, Dutch students reported feeling less connected at school, and an American study of forty thousand schools found test scores unmoved.

Florida called the money peanuts and is going to trial. New Mexico's judgment is on appeal. Nearly three thousand cases are still in the federal pile, school districts among them.

And the next feed is already here. One in five Ontario students told researchers this summer they use an AI chatbot for emotional support, and those kids were twice as likely to be in real distress. The researcher who ran it was careful to say the chatbot didn't cause that. The kids who are hurting go looking for something that answers at three in the morning, and the settlement, two hundred pages long, doesn't mention it. New Mexico's attorney general noticed that gap and said the deal lacked a ban on romantic chatbot interactions with minors. He's writing a law instead.

A teen quoted in the school-phone research asked the question that outlasts all of it: you're taking these things away, but what else do you have to offer?

Sources: ABC News, CBS News, Wired, The Verge, Washington Examiner, CBC, Fortune, TechCrunch, The Guardian, Ars Technica, the settlement agreement (D.C. Office of the Attorney General), Meta newsroom, New Mexico Department of Justice, CNBC, the Wall Street Journal (2021), HHS Surgeon General's advisory (2023), CDC Youth Risk Behavior Survey and NCHS, Nature (Odgers review, 2024; news, 2026), Platformer, BMJ, GOV.UK, France 24, JAMA Pediatrics via CBC and the New York Post, Scientific American, Harvard Gazette, LA Times, Electronic Frontier Foundation.