On the morning of Wednesday 26 August a document arrived in a federal courtroom in Oakland, California, at six o'clock. Judge Yvonne Gonzalez Rogers said she wanted to take a closer look at anything that landed at that hour. By the afternoon she had entered it as a consent judgment and every party had waived the right to appeal. The trial of Meta by four state attorneys general, opened on 18 August and calendared for six weeks, was over in its second week. The head of Instagram, Adam Mosseri, had taken the stand the day before. Mark Zuckerberg was next on the witness list. He did not get there.

The headline number was up to $18 billion. Wall Street heard it and sent Meta's shares up about one per cent by the close. That reaction is the whole story in miniature. Investors are not sentimental and they are not stupid. They looked at what the press was calling a big tobacco moment and priced it as relief.

This piece does three things. It counts the money properly. It reads the 130-page settlement agreement rather than the press releases about it. Then it asks the question that matters for anyone reading this in Ireland, which is what, if anything, the deal has to do with us. The short answer is in the agreement's own text and it is: nothing whatsoever.

Count it properly

The figures in the newspapers ranged from $16.7 billion to $18 billion because everyone was adding up different parts of the same package. The settlement agreement filed with the court settles the arithmetic. Its payment schedule, Exhibit B, totals $16,680,647,753 over ten annual instalments. That total splits into two columns. The guaranteed instalments come to $1.166 billion a year, or $11.66 billion over the decade. The contingency instalments come to $502 million a year, or $5.02 billion. Meta only pays them if its competitors sign comparable deals. More on that below. On top of the schedule sit $459,293,017.80 to close the states' Cambridge Analytica claims and $75 million for the states' legal costs. Texas negotiated its own deal for a little over $1 billion outside the coalition. Add everything and you reach roughly $18 billion. Remove the contingent column and the guaranteed teen-safety money is $1.17 billion a year.

Meta will book a charge of about $10 billion in the third quarter. That is an accounting entry for the present value of the whole liability, not cash leaving the building. The cash leaves at $1.2 billion or so a year for ten years. One-off charges are the kind of thing analysts set aside when they model a company. The same week the settlement was announced, the analyst note was that it cleared the way for new AI products.

Now put $1.17 billion a year beside Meta's own accounts. In the second quarter of 2026 alone the company recorded $2.40 billion of charges related to legal proceedings and still reported operating income of $18.78 billion for the quarter. The annual settlement payment is less than half of one quarter's legal bill and about a sixteenth of one quarter's operating profit. Full-year 2025 revenue was $200.97 billion, which is $550 million a day. The yearly payment is a little over two days of revenue. The company's capital spending guidance for 2026, almost all of it for AI, is $130 to $145 billion. In July the New York Times reported Meta was in talks to lease spare data-centre capacity to Anthropic for up to $10 billion over two years. If that deal closes, the rent on the spare room is bigger than the fine.

The states had asked for a great deal more. In a July filing Meta disclosed that the four trial states were seeking up to $1.4 trillion in penalties, a figure built by multiplying every qualifying teen user by the maximum per-violation fine in each state's consumer protection law. Meta called it a number with no analogue in the history of consumer protection enforcement. It was always a ceiling rather than a forecast. Nobody in that courtroom expected a judgment that would end the company. The gap between the ceiling and the settlement is still the gap between a trial and a deal.

Why Meta wanted the deal

The price was not the problem. The record was.

Meta had spent 2026 losing. In Santa Fe a jury found in March that the company had committed 75,000 violations of New Mexico's consumer protection law and set civil penalties at $375 million. In August the judge in that case, Bryan Biedscheid, declared Meta's platforms a public nuisance and ordered a further $567 million into an abatement fund for youth mental health, $942 million in all. The detail worth knowing is where the pollution comparison came from. It was Meta's own lawyers who argued that the court should treat the company like a polluting factory, in order to limit the remedies available. The judgment records that the court "agrees with Meta's argument" and applied the factory analogy to find a public nuisance. Meta is seeking a stay pending appeal.

In Los Angeles in March a jury returned the first personal injury verdict against Meta and Google in this wave of litigation, $6 million to a 20-year-old woman identified as KGM, who began on YouTube at six and Instagram at nine. The jury put 70 per cent of the responsibility on Meta. TikTok and Snap had settled before that trial began. Zuckerberg did testify in that case, for a day in February, visibly irritated by his own emails. So it is not true that he has never faced a jury on this subject. It is true that he did not face the states.

None of that is resolved by the August settlement. As of early August the federal docket held more than 3,000 personal injury cases and roughly 1,300 brought by school districts, none of which the deal touches. Florida and New Mexico stayed outside it. New Mexico's attorney general Raúl Torrez told Fortune the national deal was weaker than what his state had won at trial, which included a direct ban on romantic or sexualised chatbot interactions with minors and stronger safeguards against adults contacting children in private messages.

Then there was the week the Oakland trial did run. Meta's defence was that it already had adequate guardrails. The states' first witness was Arturo Béjar, who had spent six years building safety tools at Facebook and returned in 2019 after his 14-year-old daughter opened an Instagram account. "She started getting requests for sex and pictures of private parts from people that she did not know," he told the jury. Bloomberg's headline, carried by the Irish Times, was that his daughter had been prey on Instagram. A former data scientist on the wellbeing team, George Volichenko, testified that when the "Take a Break" prompt was opt-in, fewer than 0.165 per cent of teenagers turned it on and took the break. Making it default-on was considered and refused because of the effect on core metrics. He said he was told not to worry about the low numbers because the team existed "partially to protect the company against the upcoming lawsuits". The guardrails existed. Meta's own data showed almost nobody reached them.

What discovery put on the record

The deeper danger for Meta was never the witnesses. It was the paper.

A team at NYU Stern led by Jonathan Haidt and Zach Rausch has catalogued 35 internal Meta studies on whether its products harm young people, assembled from whistleblower disclosures and litigation discovery. Three of them are enough.

In 2021 Meta ran a survey it called the Bad Experiences and Encounters Framework, overseen by Béjar, across 237,923 Instagram users. Just over half, 51.6 per cent, reported at least one negative experience in the previous seven days. Among users aged 13 to 15, 13.9 per cent reported receiving unwanted sexual advances in that week. Of those advances, 93.8 per cent came from strangers. Béjar brought those numbers to the US Senate in November 2023.

In 2019 Meta ran a randomised deactivation experiment with Nielsen, Project Mercury, the design that can establish cause rather than correlation. Users who stopped using Facebook for a week reported lower depression, anxiety, loneliness and social comparison. The internal discussion, unsealed in November 2025 in the school districts' litigation, includes one employee asking whether burying the results would look like "tobacco companies doing research and knowing cigs were bad and then keeping that info to themselves". Meta halted the work and attributed the findings to the existing media narrative about the company. It says today the pilot was stopped because the method was flawed. Its own researchers wrote to Nick Clegg, then head of global affairs, that "the Nielsen study does show causal impact on social comparison".

In the same year Project Daisy tested hiding public like counts. It reduced negative social comparison, especially for teenage girls. Like counts stayed on by default because hiding them cut advertising revenue by about one per cent. One of the researchers' chat logs, now an exhibit, reads: "IG is a drug… We're basically pushers… We are causing Reward Deficit Disorder". It notes that Mosseri "freaked out when I talked about dopamine".

Set that beside what the chief executive told Congress. On 31 January 2024 Zuckerberg told the Senate Judiciary Committee that "the existing body of scientific work has not shown a causal link between using social media and young people having worse mental health outcomes". Meta had run the causal experiment five years earlier, got the answer and shelved it.

There is a further allegation that Meta denies. Sarah Wynn-Williams, its director of global public policy from 2011 to 2017, told senators in April 2025 that the company could identify when a teenager felt worthless or insecure, for example after deleting a selfie. Advertisers, she said, could then reach that teenager with beauty or weight-loss products at that moment. Meta says her account is false. What it did about her is dealt with below.

Read the carve-outs

The states got real things. A default daily limit of two hours across Instagram and Facebook that only a parent can lift. No access between midnight and 6am. Notifications muted 8am to 3pm on school days. Break prompts on by default. Like counts hidden for teenagers. Cosmetic surgery filters disabled. Age assurance with audited accuracy targets. An independent auditor, paid for by Meta and chosen jointly with a committee of states, with access to raw data and internal documents. Most of those obligations run for ten years.

Some of that was already on the shelf. Meta's Teen Accounts, announced in September 2024, had private-by-default profiles, a sleep mode muting notifications from 10pm to 7am and a reminder after 60 minutes. The settlement hardens those into a court order and adds things Meta had not volunteered: a lock rather than a reminder, a night block rather than muted notifications and an auditor. That is a fair account of what litigation bought and it is more than the "nothing new" reading allows.

The interesting part of any settlement is not what it restricts. It is what it exempts, because the exemptions tell you who held the pen. Section II.B.3 of the agreement sets the two-hour limit and then, in the same sentence, says the limit "shall not include time spent watching Longform Content, engaging in messaging, or accessing settings".

Take those in turn. The agreement's definition of a "Meta SMP", the thing being regulated, expressly excludes "direct messaging features linked to or included in those platforms, such as Instagram Direct or Messenger". WhatsApp does not appear in the document at all. Time in the settings menus is free, though the drafters did think to add that content reached through settings still counts. Then there is definition EE. "Longform Content" means "a piece of video or audio content with a duration of at least 22 minutes that Meta has determined with a high degree of reliability to not have been artificially extended". Meta decides what qualifies.

Twenty-two minutes is a very particular number. The median TikTok is well under a minute. A great deal of YouTube clears 22 minutes without trying. A limit that exempts long video and exempts direct messaging is a limit drawn around YouTube's format and Meta's messaging business and dropped onto TikTok's front lawn, since a pure short-form feed has no long-form library and no messaging line in which to hide the minutes. It also lands, as it happens, on Reels, the short-form feed Meta bolted onto Instagram to stop TikTok eating its lunch. The order restrains Meta's copy of its rival while leaving Meta's core strengths alone.

The one-hour headline deserves the same reading. Under Phase II, which only switches on if the rivals comply, the default becomes "60 minutes per day of use on each Meta SMP … not to exceed 120 minutes per day of use cumulatively across Meta SMPs". Read that twice. The combined budget across Instagram and Facebook stays at two hours. The tightening is per app.

The ratchet

Here is the mechanism that turns 52 elected prosecutors into Meta's field sales force.

The agreement names three "Core Industry Members": Snap, TikTok and YouTube. The remaining $5.02 billion is only paid once every one of them has adopted "substantively equivalent obligations" in a given state, whether by settlement, statute or verified voluntary compliance. It is also conditional on the ones that matter having paid. The payment condition applies only to Core Industry Members "with annual profits above $10 billion". Google clears that. ByteDance may. Snap does not come close. In the second quarter of 2026 Snap reported revenue of $1.6 billion, an operating loss of $171 million and a net loss of $164 million. So Snap owes the states nothing under this structure. It must simply comply with the product rules, at its own cost, to unlock money that goes to the states and stricter limits that apply to Meta.

That is the strange part. Phase II, the one-hour-per-app regime, requires "Industry-Wide Adoption" by all three named rivals. The smallest and least profitable of them holds a veto over how strict Meta's own app has to be. Meta built a settlement in which its weakest rival can block the tightening by doing nothing.

The agreement also polices what the rivals may exempt. Industry-wide adoption requires that a rival's time limit "does not exclude any features, surfaces, or portions other than messaging or Longform Content" unless Meta waives the requirement in writing. Meta may refuse only if it can show a significant competitive disadvantage. Any new platform that reaches five million US teenage monthly users averaging 30 minutes a day becomes a "New SMP Entrant" and gets five months to comply, or Meta's own obligations loosen. A safety standard priced so that only the largest incumbents can afford membership is a barrier to entry in a public health uniform.

Meta did not leave the point to inference. It published an open letter to TikTok and YouTube, then bought full-page advertisements in the New York Times, the Washington Post, the Wall Street Journal and the Los Angeles Times to run it: "We want to ensure teens benefit from this new industry standard, but we cannot do it alone." Snap, which the letter does not mention, declined to comment. Its shares fell about eight per cent in the days after the deal and Pennsylvania sued it. Nikolas Guggenberger of the University of Houston Law Center put it precisely to Engadget: the structure "fuses the interests of both the AGs and Meta together", so that both now have an incentive to pressure the other players. James Grimmelmann of Cornell noted that the rivals "have just lost Meta as an ally" in lobbying against legislation. Rob Lalka of Tulane was blunter: "Meta is not reforming here, they're complying."

Civil liberties groups read the same document and found different problems. The Electronic Frontier Foundation argues the deal embeds age assurance and surveillance into law for ten years, treats anyone who declines an age check as a teenager and gives parents and an auditor access to messaging partners and search histories. Public Knowledge's summary was that Meta has paid to write the kids' online safety rulebook for everyone else. Both can be true at once. The deal is a regulatory apparatus and it is Meta's.

In fairness to the prosecutors

Meta's strongest argument is correct as far as it goes. In the attention economy unilateral disarmament does not work. If Instagram is capped at an hour and TikTok is not, teenagers do not go outside. They go to TikTok. Meta's chief legal officer C.J. Mahoney said the framework would empower parents and that "because teens move fluidly across dozens of apps, we need an industry-wide solution". That is true. It is also exactly the argument a company would make if it wanted the regulator to raise its competitors' costs.

The attorneys general were not fooled. They could see the money was trivial to these firms, so they spent their leverage on the thing that was worth having: a supervised, audited set of product rules that cannot be quietly dropped, some of which no divided legislature would have passed. The United States Congress has spent a decade demonstrating it will not legislate on this. The alternative to the settlement was not a better settlement. It was a verdict, years of appeal and possibly nothing. A bipartisan group of professionals who had watched Meta lose twice in six months looked at the options and took what they could get now. Their tool was litigation because litigation was the only tool the system had left them.

What Meta bought was predictability. A trillion-dollar coin toss cannot be priced. A fixed, amortised line item smaller than the existing legal budget can be. The stock told you which of those Meta now has.

Housekeeping

While the chequebook was open Meta did some tidying. The $459 million for Cambridge Analytica resolves claims by 46 states and two territories over the 2018 scandal, in which data on up to 87 million people, by Facebook's own count, was improperly shared with a consultancy tied to the Trump campaign. Facebook had already paid a $5 billion federal penalty in 2019. The states' releases cover future claims based on facts not yet known. That is worth a moment. A company that knows what is in its own files bought a release for whatever else is in them.

The reflex is consistent. Faced with inconvenient facts, Meta's instinct is rarely to argue that the facts are untrue in a forum that could test them. It pays for their removal from the record, or it uses process to keep them out of one. Wynn-Williams published Careless People in March 2025. Meta did not sue her for libel, which would have required proving the book false in open court. It went to private arbitration under a non-disparagement clause in her severance agreement and, without her present, obtained an emergency order barring her from criticising the company or promoting the book, with a penalty of $50,000 for each violation, each sale counted separately. Her lawsuit to overturn the order, filed in June, alleges that for more than a year Meta sent representatives to her public appearances to photograph her and document that she said nothing. It objected to her sitting silently on a panel at the Hay Festival because the other panellists were critics. Meta's response is that she is "trying to use the legal process to sell books", that an arbitrator already found she broke her agreement and that the book is "divorced from reality". Whether or not that is so, a company accused of intrusive surveillance defended its reputation by surveilling an author to make sure she stayed quiet, and she did.

The Oakland settlement carries the same signature. Section X.C, in full: "This Agreement is entered into for settlement purposes only and does not constitute an admission by Meta of any liability, wrongdoing, or violation of any local, state, federal, or international law." The reckoning arrived. It cost two days of revenue a year. It came with a clause saying Meta admits nothing.

What it has to do with Ireland

The next sentence of Section X.C is the one that matters here. "Nothing in this Agreement shall be construed to apply to, establish a standard of care for, or serve as precedent in any non-participating U.S. state or any international jurisdiction whatsoever." When RTÉ asked whether any of the new protections would reach Irish teenagers, Meta declined to comment beyond the US terms. Instagram and Facebook are provided in Europe by Meta Platforms Ireland Limited, a Dublin company, which is why the Data Protection Commission fined Instagram €405 million in 2022 for defaulting children's accounts to public and exposing their contact details. Ireland is the home regulator for the company that has just agreed, in writing, that none of this applies to Ireland.

Two days after the Oakland trial opened, the Irish Times reported that Zuckerberg had bought Strancally Castle in Co Waterford, 440 acres, for a sum thought to be between €20 million and €30 million. His spokesperson said the family looked forward to spending time in Ireland, "where Meta maintains its international headquarters". He will use the house when here on Meta business. Good. He is welcome. This is a country that has taken in people with far less and asked nothing of them except that they keep the place decent. So it is fair to say to a new part-time neighbour what any household says to a guest: behave in front of the children. That is not a high bar. It is the most basic requirement this country places on any adult who arrives here. There are people who spend their lives insisting on it, loudly, outside hotels, for immigrants with nothing. What is good for the goose is good for the gander. If the standard is that a newcomer must not harm Irish children, then the newcomer with the castle and the company gets exactly the same standard, applied with exactly the same volume. Anything less is not a principle. It is a price point.

Two requests, in that spirit. The first concerns the children themselves. There is no Irish equivalent of Meta's internal research because only Meta can run it. What there is, is enough. In April the Commission put the share of under-13s on Instagram or Facebook across the EU at 10 to 12 per cent. On Tuesday CyberSafeKids published its survey of more than 7,000 Irish children aged 8 to 15. Seventy per cent of 8 to 12-year-olds hold accounts on platforms rated 13 and over. About a quarter of both age bands met content or unsolicited contact that bothered them, including sexual material and threats. The rates are far higher in DEIS schools. The age-assurance obligation under the Online Safety Code has been binding since July 2025 and its chief executive Alex Cooney put the state of enforcement in one sentence: "We're arguing over whether to lock the door while the laws already meant to protect children inside go unenforced." The Irish Council for Civil Liberties wrote to Meta the day after the Oakland deal asking whether it would extend the protections to children here or explain why not. No answer has been reported. The one Irish enforcement action in motion, Coimisiún na Meán's two investigations opened in May into whether Instagram and Facebook use dark patterns to steer people away from a non-profiled feed, is being challenged by Meta in the High Court on the ground that the regulator investigated the wrong Meta company. It is back before the court in October. Process over substance, again. A part-time resident of Waterford is in a position to answer the ICCL's letter himself.

The second concerns the glasses. Meta's Ray-Ban glasses film whatever the wearer is looking at, signalled by a small white light on the frame. Meta says the light cannot be switched off. For most of the product's life that was not true. In October 2025 404 Media found a hobbyist selling a $60 modification that disabled the light while leaving the glasses fully working and outwardly unchanged. A no-cost software bypass existed alongside it. According to IPVM, the surveillance-industry testing outfit, it took Meta 18 months and three firmware releases before the camera reliably stopped when the light was blocked, a fix IPVM confirmed only this month. Meta now permanently disables the camera on pairs where the light has been physically tampered with. The light, in any case, only ever protected the bystander. It never protected the wearer. In March the Swedish papers Svenska Dagbladet and Göteborgs-Posten reported that data annotators at Sama, a Meta subcontractor in Nairobi, were reviewing footage from these glasses to train Meta's AI. What they saw included people having sex, undressing, using toilets and holding bank cards. None of it was anonymised. Faces, bodies and documents were legible. The wearers had no idea a human being in another country would watch. Meta ended the Sama contract in April, saying the firm did not meet its standards, which is a way of removing the witnesses rather than the pipeline. The UK Information Commissioner has since asked Meta how the footage is processed and stored. Kenya's data protection office has opened its own investigation. Note what this is not. It is not a stranger with an antenna intercepting the glasses on the street, the version of the danger that people reach for and that Meta could refute in a sentence. The camera link is encrypted and the video never travels the way that story requires. This is the product working exactly as designed, carrying the bedroom to a reviewer in Nairobi because that is where the design sends it. The threat was never the creep with a radio. It was the company with a training set. In 2021 the Data Protection Commission and its Italian counterpart said it had not been demonstrated that any field testing had been done to show the light was an effective notice to the people being recorded. Fortune reported last month that the product has acquired the nickname "pervert glasses" for recording people undressing. More than 100,000 people have downloaded an app that warns them when a pair is nearby, the same report says. In May Senator Laura Harmon called for limits on recreational filming with them in Ireland, naming children specifically. When Zuckerberg testified in Los Angeles in February, members of his own entourage walked into a courtroom that bans recording while wearing them. Judge Carolyn Kuhl threatened them with contempt. If the chief executive's staff cannot be relied on to take them off in a courtroom, the rest of us are entitled to ask that they come off in Waterford. Ireland has enough trouble keeping strangers out of children's inboxes without cameras on the faces of the people who built the inbox.

None of this is a demand for anything he does not already have the power to give. The terms are the ones any guest gets and they are not complicated. If he wants to live peaceably in Ireland, even part of the year, he discontinues harming its children.

The Tánaiste Simon Harris called the settlement a potential "breakthrough moment" and urged every platform to adopt the protections in every jurisdiction, adding that social responsibility should not have to be compelled in court. It is a fine sentiment and it describes the exact opposite of what happened. Social responsibility was compelled in court, by 52 prosecutors with a trial in progress. The compulsion stops at the water's edge because Meta wrote the clause and the states signed it.

The European lever is a different one and it is worth understanding, because it is arguably the better lever. On 29 April the European Commission preliminarily found that Meta breaches the Digital Services Act by failing to keep under-13s off Instagram and Facebook, noting that a child can enter a false birth date with no check and that reporting an underage account takes seven clicks and usually leads nowhere. On 10 July it went further and preliminarily found the design itself in breach: infinite scroll, autoplay, push notifications and the personalised recommender were never properly assessed for their effect on the wellbeing of minors and vulnerable adults. The time-management tools, it found, "can be easily dismissed". The Commission's stated remedy is to disable autoplay and infinite scroll by default, build in real screen-time breaks and make the recommender less engagement-oriented. Fines under the Act run to six per cent of worldwide turnover. Meta says it disagrees with the findings. The day after the Oakland deal a Commission spokesperson confirmed discussions with the company aimed at "equally effective protection for our children here in the European Union". He pointed out that European commitments are expected to hold for the foreseeable future rather than for five or ten years.

Notice the difference in kind. The American settlement is a time budget with holes in it. The European demand is a design change. A two-hour clock that excludes messaging, settings and anything over 22 minutes is a clock Meta can live with, which is why it agreed to it. Switching off the machinery that makes the two hours feel like twenty minutes is the thing it did not concede in Oakland and is now being asked for in Brussels. Noeline Blackwell of the Children's Rights Alliance made the same point to RTÉ: the settlement leaves the recommendation algorithms on by default, "the very algorithms" that pull children down rabbit holes.

Irish policy is currently pulled two ways on this. In May the Oireachtas Media Committee, chaired by Alan Kelly, reported that no expert witness across its hearings had recommended an under-16 ban. It recommended instead disabling recommender algorithms for children, banning infinite scroll and restricting autoplay, which is close to the Commission's list. The Government's Digital and AI Strategy in February kept an under-16 restriction as policy while confirming that no legislation was being drafted, preferring an EU-level digital age of majority with a domestic backstop plus a State digital wallet to carry the age check. Australia's ban took effect on 10 December 2025 and removed 4.7 million under-16 accounts within a week. Ireland holds the EU presidency for the second half of this year and has named the question a priority. The Oakland document is a useful exhibit for that debate, not as a model, since it explicitly declines to be one, nor as a threat, since the money is trivial. It is a demonstration of how a time limit gets drafted when the company being limited is in the room.

Coda

One last asymmetry. While the entire legal apparatus of the United States concentrated on teenagers who spend an hour or two a day on Instagram, the Economist reported last October that pensioners now spend more than half their waking hours looking at screens and are more susceptible than the young to hoaxes and scams. Nobody has filed a $1.4 trillion suit on behalf of granddad. Partly that is because an adult is entitled to waste an afternoon. Partly it is because a misled pensioner, with a live card on the iPad and a vote in the ballot box, is useful to more people than a protected one.

Aaron Sorkin's The Social Reckoning, dramatising the Facebook Files and the whistleblower disclosures that fed this litigation, opens on 9 October. The reckoning was supposed to take place in a federal courtroom in Oakland, where the internal research could be tested against evidence and made fact. It was packaged into an accounting reserve instead, before the chief executive answered a question. So the public will watch it in a cinema.

Sources

Note on AI use: I used an AI system as a research and drafting assistant on this piece. I directed the work. I checked every factual claim in it against the primary sources listed above. I decided what went in and what stayed out. The judgements here are mine. So are the errors. So is the responsibility for publishing it. This disclosure is made in line with Article 50(4) of the EU AI Act, whose transparency obligations have applied since 2 August 2026.