Only partly. Eight jurors were seated in the Oakland federal court, but their verdict is advisory. US District Judge Yvonne Gonzalez Rogers determines whether Meta is liable, and if she finds against the company she can impose civil penalties and order changes to Facebook and Instagram.
How much could Meta have to pay in the 29-state trial?
The two sides are far apart. Meta told the court civil penalties could reach roughly $1.4 trillion, close to its entire market value. The states put the figure nearer $200 billion, about three years of after-tax profit. Both are arguments made in court rather than determinations, and the judge has not set any figure.
Why are only four states presenting the case?
California, Colorado, Kentucky and New Jersey are trying the case first as lead states in a bipartisan coalition of 29. It is a bellwether structure: the lead group tests the shared legal theory, and the outcome shapes the leverage and settlement posture of the remaining states.
What are the states accusing Meta of?
The states allege Facebook and Instagram were designed to maximise engagement among minors in ways that harmed their mental health, and that Meta's public statements about platform safety did not match what it understood internally. Meta disputes the claims, arguing the states misrepresent its platforms and policies and selectively quoted internal documents.
Saba Javed covers the point where AI products run into security problems and the law. Recent stories include researchers tying OpenAI agents to a RubyGems attack, Claude Code switching to auto mode by default despite an 11% miss rate in testing, and the FTC's position that complying with state AI laws can still be illegal.
01The Jury Is Advisory-Eight jurors were seated, but their verdict only advises. US District Judge Yvonne Gonzalez Rogers alone determines Meta liability, which makes this a documents case rather than a sympathy case.
02The Numbers Are Advocacy-Meta told the court penalties could reach about $1.4 trillion, close to its market value. The states put it nearer $200 billion, roughly three years of after-tax profit. Both are positioning, not forecasts.
03The Remedy Is The Real Risk-Rogers can order changes to Facebook and Instagram. Meta quantified the possible fine and did not quantify the remedy, because court-ordered design changes to engagement systems cannot be modelled in advance.
04It Is A Bellwether-Four of the 29 states are trying the case first. A liability finding gives the other twenty-five a proven template and substantially more settlement leverage.
The number in every headline out of Oakland this week is $1.4 trillion. It is the wrong number to be watching, and Meta is the one who put it there.
A bipartisan coalition of 29 states began trying its case against Meta on Tuesday in federal court in Oakland, arguing that Facebook and Instagram were deliberately designed to keep young people engaged past the point of harm, and that the company misled the public about it. Lawyers for California, Colorado, Kentucky and New Jersey are running the case first, with the remaining states behind them. It was filed in 2023.
Meta told the court that civil penalties could reach roughly $1.4 trillion, which is approximately the company's entire market value. The states put the figure closer to $200 billion, or about three years of after-tax profit. Both numbers are advocacy. Only one of them is designed to be quoted.
The structural detail almost nobody is leading with
Eight jurors were selected. Their verdict is advisory.
US District Judge Yvonne Gonzalez Rogers determines Meta's liability, and if she finds against the company she can impose civil penalties and order changes to Facebook and Instagram. The jury's role here is to advise, not to decide, which is a meaningfully different proceeding from the one most coverage is describing.
That distinction matters for how you read everything that follows over the coming weeks. Testimony is being staged for a jury whose conclusion is not binding, in front of a judge who will reach her own. Dramatic moments will be reported as though they move a needle, and the needle they move is not the one connected to the outcome. When Mark Zuckerberg testifies, as he is expected to, the coverage will treat it as a jury moment. It is a Rogers moment.
It also means the usual corporate defence calculus is inverted. A company facing a jury can play for sympathy, confusion, or a single holdout. A company facing a federal judge on a documentary record built over three years has to win on the documents.
Why Meta's own number is the tell
There is something odd about a defendant volunteering that it might owe $1.4 trillion.
Companies do not usually publicise the ceiling of their own exposure. Meta did it because the figure functions as an argument rather than an estimate. A penalty framework that can produce a number equal to a defendant's entire market capitalisation invites the response that the framework is not being applied sensibly. It is a way of saying the statute yields absurd results at scale, without having to say so directly.
The states' $200 billion does similar work in the opposite direction. Anchored against three years of after-tax profit, it is built to sound survivable, proportionate, and therefore reasonable to impose. Neither figure is a forecast. Both are positioning for the moment Rogers has to pick a number, and the gap between them is the negotiating range in public view.
Which is why the money is not the interesting part.
The remedy is the exposure
Rogers can order changes to Facebook and Instagram. That sentence is doing more work than the trillion-dollar one.
A financial penalty, even a very large one, is a known quantity. It is paid, it is written off, the share price absorbs it, and the business continues. Court-ordered design changes to the recommendation and engagement systems of two platforms with billions of users are not a known quantity. They are permanent, they are hard to reverse, they set a template other regulators will copy, and their cost cannot be modelled in advance because it depends on how much of the product's economics rest on the mechanics being changed.
If you want to know which outcome Meta fears, look at which one it quantified. The company gave the court a number for the fine. It did not give the court a number for the remedy, because there is not one.
What each side is actually arguing
The states' framing is compact enough to survive a long trial. In her opening, California Deputy Attorney General Megan O'Neill described a business model in four verbs: hook users, hold them as long as possible, harvest their data, and hide the truth in public statements. The claim is that engagement time among minors was a revenue input, that the company understood the mental health consequences internally, and that its public safety messaging did not match what it knew.
Meta's counsel argued that the states misrepresent the platforms and their policies, that the case lacks supporting evidence, and that the attorneys general cherrypicked internal documents and employee comments to construct a narrative. That defence is neither unusual nor unreasonable. Large companies generate enormous volumes of internal dissent, and any sufficiently large document set contains someone arguing the alarming version of any proposition.
The question Rogers has to answer is not whether alarming internal documents exist. In a company of Meta's size, they always do. It is whether the documents describe something the company then acted on, and whether the public statements made at the same time were consistent with them.
There is precedent worth noting, though it is smaller and in a different forum. In March, a Los Angeles jury ordered Meta and Google to pay $6 million to a woman who said she became addicted to Instagram and YouTube as a child. Financially that is a rounding error against either figure being argued in Oakland. As a signal about how a lay factfinder responds to this category of evidence, presented in this way, it is not nothing, and it arrives alongside a growing body of private suits running in parallel to the state actions. Meta won few of the arguments it will need to win again here.
Why only four states are in the room
Twenty-nine states signed on. Four are trying the case. That is not a scheduling accident, it is the structure of a sprawling multidistrict fight in which a lead group tests a shared legal theory and everyone else calibrates against the result.
It makes this a bellwether in the proper sense. The four states are not litigating only their own consumer protection claims. They are establishing whether the theory survives contact with a federal courtroom, a complete documentary record and Meta's counsel. If it does, the remaining twenty-five arrive with a proven template and considerably more leverage. If it does not, the coalition's position weakens everywhere at once, including in the private cases.
Bellwethers also compress information. Everything the lead states learn about which arguments land with Rogers becomes a shared asset for the others. Everything Meta learns about which defences work becomes one too. This trial is a discovery process for both sides about the next several years of litigation, which is part of why neither is behaving as though a single penalty figure is the real stake.
Why the timing is awkward for the industry
This trial opens into a season of teen safety product announcements across the sector. OpenAI has been building out a distinct experience for younger users, following earlier work on teen alerting and activity restrictions in ChatGPT. Most large platforms now have some version of the same roadmap.
Those launches are defensible on their own terms. They also create an evidentiary problem that companies rarely think about when shipping them, because a safety feature is an implicit statement that a risk exists and was addressable. Ship it in 2026 and the obvious question is what was known in 2021 and why the mitigation took five years.
This is the same dynamic visible in state-level enforcement against AI products, where the gap between what a company's own risk assessment says internally and what its marketing says externally is doing most of the legal work. It is also why published internal risk evaluations have become such a delicate exercise. Documenting a hazard honestly is good practice and creates a record. Not documenting it is worse practice and creates a different record.
There is no clean way out of that bind. The only durable position is for the internal assessment and the public statement to say the same thing, which is precisely what the states allege did not happen here.
What to watch
Three things, none of which is the damages number.
The documents Rogers engages with. In a bench-decided liability finding, the judge's questions during testimony are the real signal. Which internal exhibits she returns to will tell you more than any single day of testimony.
Whether the remedy phase gets separated. If liability and remedy are split, the design-change question becomes its own fight with its own record, and that is where the durable consequences for the products get decided.
What the remaining 25 states do. Four states are trying this case. The other twenty-five are watching a live test of the same theory. A liability finding here does not automatically resolve their claims, but it substantially changes the settlement posture of every one of them.
The trillion-dollar figure will lead the coverage for weeks. It is the least informative thing in the courtroom.
Legal and policy analysis, not investment or legal advice. The allegations described here are claims made by the plaintiff states and have not been proven; Meta disputes them. Penalty figures are the estimates each side argued in court in August 2026, not determinations by the court.
Saba Javed covers the point where AI products run into security problems and the law. Recent stories include researchers tying OpenAI agents to a RubyGems attack, Claude Code switching to auto mode by default despite an 11% miss rate in testing, and the FTC's position that complying with state AI laws can still be illegal.
Comments