For about four hours, seventeen billion dollars sounded to me like justice.
That is roughly how long it took to get the filed document open and start reading, and I want to be honest about the order those two things happened in, because the number did exactly what a number that size is supposed to do. It landed, it felt proportionate to a decade of accusations, and some part of my brain filed the story under resolved before I had read a line of it.
Then you read the agreement and the feeling reorganizes itself. Not into outrage — this is not a scandal, and the people who negotiated it are not fools. Into something more uncomfortable: the recognition that the biggest number in the history of this fight is doing less work than the sentences around it, and that the sentences around it were written by the defendant.
What is actually in the document
Start with the money, because it is the part everyone repeated and almost nobody checked.
The filed consent judgment and the executed agreement behind it run to about 130 pages, and Exhibit B lays the payments out jurisdiction by jurisdiction. Meta "shall make payments to each Settling State in ten installments." The maximum across all of them, plus a separate Cambridge Analytica component, comes to just over $17.1 billion. But the guaranteed floor — the part Meta is actually obliged to pay regardless of what anyone else does — is $12.19 billion, as Connecticut's attorney general states plainly. California's office uses the same careful construction: up to $17 billion to the states over ten years.
Hold onto that phrase. Up to. It is doing an enormous amount of quiet work.
You will also see $18 billion in a lot of coverage. That figure is Meta's own characterization of its agreement; CNBC, which reports the filed maximum as $16.7 billion, quotes the company saying the deal "includes a payment of approximately $18 billion." Nothing in Exhibit B supports it. When the defendant's press number runs about five percent above the $17.1 billion maximum — and nearly eight percent above Exhibit B itself — that is worth noticing on its own.
Now the denominator. Meta's audited results put full-year 2025 net income at $60.458 billion. So the ceiling is a little over a quarter of a single year's profit. The guaranteed floor is about a fifth. And because it arrives in ten annual pieces, each installment lands between about 1.9 and 2.8 percent of one year's earnings, depending on whether the contingent portion is ever paid. Fortune ran the same arithmetic against revenue and got a cleaner line: spread across a decade it is roughly 1% of Meta's expected revenue over the same period.
The design terms are the substantive half, and several of them are genuinely good. NPR's summary captures the shape: a default two-hour daily limit for under-eighteens, a nighttime block from midnight to 6 a.m., both liftable only by a parent, plus notification blocks during school hours, no visible like counts on minors' posts, and no cosmetic-surgery filters. In the agreement's own words, teen accounts default to "a daily limit maximum of use cumulatively across all Meta SMPs of 2 hours per day."
One correction that matters if you are counting on this. The agreement term is ten years — but the flagship time limits are not. Connecticut's release says it in one sentence: "These limits remain in effect for five years." Most of the deal runs a decade; the two-hour cap and the midnight blackout run half that.
The clause nobody put in a headline
Here is where it turns.
The consent judgment names one defendant. "The Defendant is Meta Platforms, Inc." The obligations bind "the Settling States, Meta, and any of Meta's respective successors, assigns or other entities or persons otherwise bound by law." That is the whole list. And to remove any doubt about who else might be reached: nothing in the agreement "shall create or give rise to a private right of action of any kind or create any right in a non-party to enforce any aspect" of it.
Snap, TikTok and YouTube do appear — by name. The agreement defines them as "Core Industry Members," and defines "Industry-Wide Adoption" as the state of affairs in which all three have "entered into a binding settlement agreement, consent decree, or other enforceable commitment" imposing "substantively equivalent obligations." Read that construction carefully. It is not a rule binding them. It is a description of the thing they would have to do voluntarily, written into a document they never signed.
Why does it appear at all? Because it is a payment trigger. Roughly $5 billion of the headline figure — about 30%, as TIME breaks it out — is contingent on the rivals signing up. And if they never do, the agreement is explicit about where that money goes: "the Contingency Installment Payments shall be permanently forfeited by such Settling State and retained by Meta."
Retained by Meta. Not held, not redirected to another remedy. If Meta's competitors decline to adopt Meta's rulebook, Meta keeps nearly a third of its own settlement.
There is a second, tighter regime — Phase II, with a one-hour per-app limit — that switches on only if Industry-Wide Adoption happens. So the strictest protections for teenagers, in a deal justified by protecting teenagers, exist in a conditional clause whose condition is controlled by companies that are not parties to it.
And Meta is campaigning for that condition openly. Its chief legal officer said the new commitments "set the right path forward for our whole industry, but this framework will only work if all our peers join us," and called on TikTok and YouTube to implement it "right away." Fortune's read is the unsentimental one: Meta has structured part of its own penalty as a recruitment device.
The strongest case for the deal, which is not weak
I want to put the other side properly, because there is a serious version of it and I have not seen it made well.
The states were mid-trial. Adam Mosseri was on the stand, Zuckerberg was expected, and everyone in that courtroom knew what the alternative looked like: a verdict, then appeals, then more appeals, and design changes arriving — if they arrived — somewhere around 2032. What the states took instead was money now and product changes on a defined clock. Judge Yvonne Gonzalez Rogers, who did not simply rubber-stamp it, called it "an excellent way to resolve" the case before signing it later the same day.
The changes themselves are not cosmetic. Default settings are the most powerful regulatory instrument anyone has over consumer software, because almost nobody changes them — which is why the industry fought over defaults so hard in the first place. A default two-hour cap on teenage accounts across Instagram and Facebook is a larger intervention than any bill Congress has managed to pass.
And there is a genuine dissent from the other direction that deserves airing. New Mexico's attorney general, the one prosecutor who took Meta all the way to a verdict, said the national deal was weaker than what his state won in court — it lacks "a direct ban on romantic and sexualized AI chatbot interactions with minors" — and then said it "still represents real progress." That is a man with every incentive to trash it declining to.
So: real money, real defaults, real speed. If you think the alternative was a better outcome, you have to explain how you get there through six years of appellate litigation with the same teenagers aging out of the class the entire time.
Here is what I still cannot get past. Every one of those defaults depends on knowing who is a teenager, and the agreement requires Meta to apply age assurance "to each Meta SMP user in the Settling States" — everybody, not just minors. That machinery is the load-bearing element the whole deal rests on, and it does not work well yet. A Syracuse researcher put it bluntly: the changes "seem great on paper, but they are all reliant upon successful, effective, and unbiased age verification technology," and "the way that we do that right now is not really effective." The federal government's own testing agrees — NIST's evaluation of age-estimation software found "a wide range in performance," and error rates "almost always higher for female faces than for males."
Meanwhile, Australia did it the other way — and it is not going smoothly either
The comparison writes itself, and the timing is almost absurd: while Meta's consent judgment was two weeks old, Australia's Parliament passed its enforcement-strengthening bill on September 10, 2026 — expanding the eSafety Commissioner's information-gathering powers and raising civil penalties for non-compliance.
Australia's under-16 rule is not a settlement. It is an amendment to the Online Safety Act, Act No. 127 of 2024, passed by a legislature, binding every covered platform, enforced by a regulator with penalty powers. Nobody negotiated its terms with the company most affected by them. And the difference shows up in the most telling place imaginable: Meta itself complies with it, and publishes what it did — more than 750,000 accounts removed in Australia as of June 30, 2026, identified partly by AI reading posts and bios for "contextual clues" like birthday celebrations or mentions of school grades.
That is the structural contrast in one image. In Canberra, Meta follows somebody else's rule. In Oakland, Meta wrote the rule and gets a refund if its competitors decline to follow it.
Now the part that keeps this from being a commercial for Australia, because it is not working the way its advocates hoped. Three months in, more than eight in ten Australian teenagers were still using social media, with platforms' failure to implement effective age checks cited as the main reason. Account ownership fell from 52% to 42%; daily use barely moved, and around half the children who kept their accounts said platforms never checked their age. The same research flagged an unintended consequence worth sitting with — a decline in parental awareness of what children were doing online.
It is contested on principle, too. Index on Censorship calls a blanket ban a disproportionate threat to free expression, and notes that enforcing it means users "upload sensitive documents such as a passport or bank card to platforms, or consent to face scanning, run by untested and opaque third-party providers." Two fifteen-year-olds took the law to the High Court, arguing it burdens the implied freedom of political communication.
So this is not a story about one country getting it right. Both approaches are struggling with the same intractable problem, which is that you cannot enforce an age rule without an age-detection system nobody has built well. The difference is not effectiveness. It is authorship — who wrote the rule, who can change it, and who can be held to it.
Run it forward
Say the rivals do sign. Phase II switches on, the one-hour limit arrives, and by 2028 the American teenage internet is governed by terms one company negotiated with fifty-one attorneys general and no legislature ever voted on. It will probably be better than what exists now. It will also be effectively unamendable — you cannot lobby a consent judgment, you cannot vote out its terms, and as Public Knowledge notes, a settlement leaves the users actually affected with no clear way to contest terms that would draw First Amendment challenges as legislation.
Now say they do not sign. Meta keeps its $5 billion, the tighter protections never activate, and the settlement's youth-safety ceiling is set permanently at the level Meta agreed to in the first place. Either branch ends somewhere Meta chose.
And then the thing nobody is planning for: a company with four employees builds something teenagers like. It now faces a compliance architecture designed around a firm with thousands of trust-and-safety staff. Techdirt's description is unkind and hard to refute — a standard only a giant can meet, where a six-hour turnaround on most reports "is achievable when you have thousands of trust & safety staff and a decade of tooling. For a startup with four employees and a Discord server, it's a fantasy."
What the people who watch this closely are saying
The alignment here is genuinely strange, and worth your attention for that reason alone.
From the digital-rights left, the Electronic Frontier Foundation argues the remedy points the wrong way entirely: the settlement requires Meta to collect, analyze, and retain more information about its teen users "when the pressure should have been on Meta to diminish its surveillance capitalism."
From the libertarian right, the same structural read with a different accent. Reason's Liz Wolfe argues that whatever you make of the child-harm case, what should unite libertarians is Meta "using the levers of the law to screw over its competitors" — and quotes the tech-industry advocate Adam Kovacevich, posting on X, calling the deal less a child-safety win than "a prosecutorial gun aimed at TikTok, Snap, and YouTube."
Child-safety advocates, who wanted this fight and largely won it, are not celebrating cleanly either. Fairplay called the settlement "a significant step," then immediately said what it does not do: "We are disappointed that the settlement does not turn off by default recommendation algorithms." The mechanism most people blame for the harm was left running.
And the process objection, from the center: Samir Jain of the Center for Democracy and Technology told NPR this is "an unusual approach to developing regulation" — that normally "we have processes in place, whether that's passing laws in Congress or an agency exercising its regulatory power," involving safeguards "including making sure laws or rules don't violate constitutional rights."
When groups this far apart describe the same document in the same structural terms, the disagreement is not really about children. It is about who gets to write rules.
What does this mean for you?
If you have a teenager on Instagram, something real changes — check it yourself. Defaults are being reset, not offered. Look at the account's time limit and night settings directly rather than trusting a press release, and know that only a parent can lift them.
Expect to prove your own age, whoever you are. The agreement applies age assurance to every user in the settling states. If you are forty-five and have never uploaded ID to a social network, that is the change coming for you, and it arrived without anyone voting on it.
Watch what Snap, TikTok and YouTube do, and know what is riding on it. If they refuse, Meta keeps roughly $5 billion and the tighter limits never switch on. That is an unusual set of incentives to have quietly attached to your children's screen time.
Learn the difference between a law and a settlement, because it is about to matter a lot. A statute can be amended, challenged, and repealed by people you elect. A consent judgment binds the parties who signed it and nobody else, and the affected public is not among them.
Do not let the number do your thinking. "Up to $17 billion" is not $17 billion, ten installments are not a lump sum, and a fifth of one year's profit is a real cost that is nonetheless survivable by design. Every one of those distinctions was in the filed document and out of the headline.
The rulebook and who holds the pen
I keep returning to the forfeiture clause, because it is the one thing in this deal that could not have been an accident.
Somebody drafted it. Somebody on the other side read it and agreed. And what it says, stripped of the defined terms, is that a company accused of harming children negotiated an arrangement in which nearly a third of the penalty comes back to that company if the rest of the industry declines to adopt the standards it wrote. You can call that clever lawyering and you would be right. You can also call it the clearest possible statement of who was setting terms in that room.
The thing I would want, and the thing nobody in this story was able to deliver, is dull: an actual law, debated in public, applying to everyone in the market, amendable when it turns out to be wrong. Australia has one and is discovering how hard the enforcement is. America now has a rulebook instead, which will probably work better in the short run and cannot be corrected by anyone who did not sign it.
Meta denies liability, and the agreement says so in terms. What it does not deny — what it quite deliberately asserts — is authorship. My vote, for whatever it is worth: notice who is holding the pen before you admire the size of the check.
The HAIA Foundation reads the filed document rather than the press release, because the two are more different than they should be. If that is a habit you want more of, it is all over here, and it is free.