Meta Pays 567 Million Dollars: The Feed Stays Exactly the Same

Paragraph 173 of the order runs to one sentence. “The Court grants no abatement relief relative to the design and implementation of Meta's algorithms.”

It sits on page sixty of a sixty-eight-page document filed at 4.41pm on 6 August 2026 in the First Judicial District Court in Santa Fe County, cause number D-101-CV-2023-02838, State of New Mexico ex rel. Raúl Torrez, Attorney General v. Meta Platforms, Inc. Fourteen pages earlier, the same judge, Chief District Court Judge Bryan Biedscheid, had ordered Meta to deposit 567 million dollars into an abatement fund, four hundred and twenty million of it earmarked for treatment services for young people. Thirty pages earlier he had written that “the weight of the evidence presented demonstrates that Meta's platforms are a cause of and substantial contributing factor to the youth mental health crisis in New Mexico”. Twelve pages earlier still, he had rejected Meta's defence under Section 230 of the Communications Decency Act on the ground that the state's claim targeted “various platform designs and features, rather than third-party content”.

Read those four passages in sequence and the shape of the ruling becomes clear, and slightly vertiginous. The design caused the harm. The design was Meta's own conduct rather than somebody else's speech. Meta must pay for the consequences. And the court declined to order any change to the design.

This is the first time an American court has put a price on the mental health consequences of an engagement-optimising recommendation system and ordered the company that built it to pay. Attorney General Raúl Torrez called the judgment “a blueprint”, and told the public radio station KUNM that “if this judgment were replicated in all fifty states, it would be collectively 150 billion dollars in liability for Meta”. That arithmetic scales New Mexico's two million residents against a national population of about 340 million. It is a thought experiment about a hypothetical fifty-state judgment, not a description of what happened in Santa Fe.

What happened in Santa Fe is narrower and more interesting. A court examined an optimisation system, concluded it was making children measurably worse, and then discovered that almost none of the tools available to a court of equity could reach it. What it could reach was money, plus a list of peripheral product constraints. Whether that alters the objective function that produced the harm is a question the ruling itself, read closely, answers in the negative.

What the Court Found and What It Declined to Touch

The case ran in two phases. A jury sat from 2 February to 24 March 2026 on the state's Unfair Practices Act claims and returned 375 million dollars in civil penalties, calculated at the statutory maximum of 5,000 dollars across roughly 75,000 violations. A bench trial then ran from 4 to 22 May on the public nuisance theory and the abatement relief that flows from it. The August order is the product of that second phase, and brings Meta's total New Mexico liability to 942 million dollars.

The factual findings are unusually specific for this genre of litigation. Drawing on the state's expert Zachary Ward, who analysed New Mexico Youth Risk and Resiliency Survey data, the court found that of New Mexico children aged eleven to seventeen with at least one mental health concern, twenty-one per cent would not have had it but for social media use. It recorded Ward's component estimates verbatim: twenty-six per cent of those with depression symptoms, twenty per cent of those with anxiety symptoms, fifteen per cent of those sleeping less than eight hours and twenty-two per cent of those carrying at least one suicide risk factor would not have done so absent social media exposure. Meta's experts, including Laurence Steinberg, Randy Auerbach and Michael Baiocchi, argued that the studies do not definitively establish causation. The court found their analysis “restrictive” and inconsistent with the trial record.

On the mechanism, paragraph 161 is blunt. “Regardless of whether it is labelled as an 'addiction' or 'problematic use', the evidence at trial proved that design elements, such as autoplay, infinite scroll, 'like counts', and push notifications create a product that, particularly for adolescent users, is highly rewarding psychologically and neurochemically.” Paragraph 162 adds that “algorithmic content recommendations can create harmful feedback loops and 'rabbit hole' users”. The court noted internal Meta data showing the top one per cent of New Mexico teenagers by time spent on Facebook were on that platform alone for at least 290 minutes a day, close to five hours.

Then came the refusals. The court declined to restrict autoplay and infinite scroll, reasoning that “regulation of industry-wide features that contribute to problematic use... require legislative or executive branch action”, that ordering changes “would risk competitively damaging Meta's platforms to an unreasonable extent”, and that such an order might run “afoul of the First Amendment and Section 230”. It declined algorithmic relief outright, holding the state's proposals “vague and aspirational, rather than objective and measurable, and would unreasonably interfere with the functioning of Meta's business activities (equivalent to putting the factory in receivership in order to abate its pollution)”.

The factory analogy was Meta's. The court adopted it, and adopted its logic with it. Pollution can be abated. The production line cannot be redesigned by a judge.

Anderson Cuts Both Ways and Meta Needed Only One Edge

The most consequential move in the order is a doctrinal pincer the court walked into without appearing to notice it had closed.

To defeat Meta's Section 230 defence, Biedscheid relied on three cases: Lemmon v. Snap in the Ninth Circuit in 2021, the Nevada Supreme Court's 2025 decision in TikTok, Inc. v. Eighth Judicial District Court, and, most importantly, the Third Circuit's August 2024 decision in Anderson v. TikTok. Anderson arose from the death of ten-year-old Nylah Anderson, who attempted a “Blackout Challenge” video that TikTok's algorithm had served her. The Third Circuit held that Section 230 did not immunise TikTok, because the recommendation algorithm was “the platform's own expressive activity” and Section 230 protects only against liability for third-party content.

That holding is a gift to plaintiffs, and New Mexico used it. But it rests on a premise with a sharp reverse edge. The Third Circuit reasoned from the Supreme Court's 2024 decision in Moody v. NetChoice, which treated a platform's compilation and curation of content as its own protected expression. If algorithmic curation is the platform's speech for First Amendment purposes, it must be the platform's speech for Section 230 purposes too. The two doctrines are inversely related. Losing immunity means gaining constitutional protection.

Biedscheid took the first half and then, forty pages later, ran into the second. Paragraph 172 states that the requested algorithm relief “would directly alter and limit the manner in which platform content is displayed. As a result, Plaintiff's requests would likely violate the First Amendment and Section 230.” Paragraph 163 explains why: autoplay, infinite scroll and algorithmic recommendations “are most closely tied to content presentation”.

So the same characteristic of the recommendation system — that it is Meta's own editorial act rather than a neutral conduit — establishes liability and then insulates the remedy. The algorithm is the company's conduct when the question is who pays. It is the company's speech when the question is whether it must change. Meta lost the immunity argument and won the injunction argument on identical grounds, and paragraph 178 concedes as much, describing the disclosure remedies the court did order as necessary precisely to mitigate the harms “that the First Amendment, Section 230, and this Court's jurisdiction prevent from being more directly addressed in abatement relief”.

That is not judicial timidity. It is a structural feature of American law as it currently stands, and it means that liability for algorithmic harm and control over algorithmic design have been decoupled at the point where they most needed to meet.

Twenty Hours of Revenue, Already Provisioned

Now the money, which is where the “cost of doing business” objection either lands or does not.

Meta reported second-quarter 2026 revenue of 60.801 billion dollars, up twenty-eight per cent year on year, and net income of 15.848 billion. That is a revenue run rate of roughly 668 million dollars a day. The 567 million dollar abatement fund is therefore about twenty hours of company revenue. The full 942 million dollars of New Mexico liability is about thirty-four hours. Measured against full-year 2025 figures — revenue of 200.97 billion dollars and net income of 60.46 billion — the abatement fund equals roughly three and a half days of profit. Meta closed the second quarter of 2026 holding 90.26 billion dollars in cash and marketable securities. The fund is 0.6 per cent of that balance.

The more telling number is on the expense line. In the same quarter, reported on 29 July 2026, a week before the Santa Fe order landed, Meta booked 2.4 billion dollars in charges “related to legal proceedings” — the single largest driver of the quarter's cost increase, large enough to prompt the company to raise the low end of its full-year expense guidance to 165 billion dollars. Meta did not itemise the charge, so the inference has to be drawn rather than read off. It does not have to be drawn unaided. On the same quarter's earnings call, chief financial officer Susan Li told analysts that the company continues to see scrutiny on youth-related issues in several markets and has a number of youth-related trials scheduled in the United States this year which may ultimately result in a material loss. Meta has separately warned investors that its exposure on claims that its platforms were designed to be addictive could run to the high tens of billions of dollars. A company that tells its shareholders to expect a material loss from youth litigation, in the quarter it books 2.4 billion dollars of legal charges, a week before a youth litigation judgment lands, has not been surprised by the judgment. The 567 million dollars is not a shock to the accounts. It is a provision being drawn down.

The historical record on penalties of this shape is not encouraging. In July 2019 the Federal Trade Commission imposed a five billion dollar penalty on Facebook over the Cambridge Analytica affair, then the largest privacy fine levied anywhere by a factor of nearly twenty. When it was reported, Facebook's share price rose, adding roughly ten billion dollars to its market capitalisation — double the fine. In July 2024 Texas secured a 1.4 billion dollar biometric settlement, payable over five years. In April 2025 the European Commission fined Meta two hundred million euros under the Digital Markets Act over its “pay or consent” advertising model.

Only one of those changed the product, and it was not the one with the biggest number. Meta stopped using facial recognition in 2021, before the Texas settlement. The DMA fine was trivial in cash terms, but the accompanying non-compliance decision forced Meta to offer European users a genuinely less-personalised advertising option from January 2026. The lesson is not that penalties never work. It is that penalties work when they are attached to a specification of what the product must do differently, and function as an ordinary operating expense when they are not.

The Harm Was Priced at Billions and Billed at Millions

The abatement figure itself repays attention, because the court showed its working and the working concedes something important.

The state asked for 953 million dollars. Its abatement plan, prepared by Dana Weiner and costed by the economist Kelly O'Donnelly, proposed a fifteen-year programme; the state's experts put the cost of fully addressing the harm statewide at around 3.7 billion dollars. Biedscheid cut the drawdown period to five years, on the reasoning that “a 15-year plan would necessarily involve treating harms beyond those that exist today”. He declined to fund construction of new clinics or hospitals as beyond what is needed “to abate current harm”. And, repeatedly and explicitly, he reduced the allocations “based on a reduced time period and Meta's market share”, noting that “other social media companies share responsibility for harm to New Mexico teenagers” but were not before the court.

The resulting allocation is itemised: four hundred and twenty million for treatment, ninety million for screening and assessment, thirty-three million for awareness and prevention, fifteen million for referral and linkage, nine million for implementation and evaluation.

Every one of those reductions is defensible on its own terms. Together they produce a figure representing Meta's apportioned share of five years of the current stock of damage. It is a settlement of accrued harm, not a charge on the activity that produces it. Nothing in the calculation is indexed to future conduct. If Meta's platforms generate the same attributable fractions of adolescent depression, anxiety and sleep deprivation over the next five years as they did over the last five, the company owes exactly the same 567 million dollars, because that number was fixed by reference to harm already done.

There is also a timing problem the order creates for itself. Paragraph 131 provides that the five-year abatement period “shall be tolled during the pendency of any appeals, provided that Meta pays a supersedeas bond”. Meta has said it will appeal, and a company with ninety billion dollars in liquid assets will not struggle to post a bond. The clock on every behavioural obligation in the order can therefore be stopped by a filing and a surety, for however long New Mexico's appellate courts take.

Meta's public response to the ruling was short. “We disagree with the ruling and will appeal,” said Andy Stone, the company's communications chief. Meta added that it works hard to keep people safe on its platforms, that it has been transparent about the challenges of identifying and removing bad actors and harmful content, that it remains “confident in our record of protecting teens online”, and that it will “continue to defend ourselves against claims that misrepresent the facts”. The appeal to the New Mexico Court of Appeals rests on First Amendment and Section 230 grounds, the two doctrines the trial court accepted as limits on the remedy having rejected them as a bar to liability, and Meta is separately appealing the 375 million dollar first-phase jury verdict. Neither appeal obliges the company to change anything about how content is ranked while it proceeds.

Push Notifications, Like Counts and Ninety Hours a Month

The injunctive half of the ruling is where the interesting engineering lives, and it deserves to be judged on its merits rather than dismissed as consolation.

Meta must eliminate push notifications to accounts of known or estimated under-eighteens in New Mexico between 10pm and 7am every day, and between 8am and 3pm on school days during term, with carve-outs for messages from connected users and for security or hazard alerts. It must hide “like counts” by default on all under-eighteen accounts, overridable only with a parent or guardian's permission. It must impose a mandatory usage cap of ninety hours a month cumulatively across Facebook and Instagram for under-eighteens — roughly three hours a day.

It must proactively seek proof of age from accounts predicted to belong to under-thirteens, delete unverified accounts after thirty days, and commit to “attempting to develop, within two years, a dedicated under-13-years-of-age prediction model (using reasonable best efforts in light of COPPA limitations)” — the Children's Online Privacy Protection Act being, as the order notes, itself a limit on what age verification a court may compel. It must bar recommendations of under-eighteen accounts to unconnected adults, bar unconnected adults from messaging minors, and prohibit under-eighteens from romantic or sexualised interactions with chatbots. On child sexual abuse material, it must report a baseline detection rate to the National Center for Missing and Exploited Children and the state within thirty days, and improve on that baseline by no less than five per cent over the abatement period.

Note the logic separating what was ordered from what was refused. Paragraph 165 is explicit: push notifications and like counts are ordered precisely because they are “least connected with platform content”. Like counts are “merely a feature created and offered by Meta to principally track and motivate user feedback, all without altering the underlying published content”. The court could reach them because they are not speech.

This is a real and defensible line, and some of these measures will bite. A ninety-hour cap is a genuine constraint on the metric the entire business optimises, and an under-thirteen prediction model is a substantial engineering commitment. But look at what the boundary produces. The court could regulate the delivery mechanism, the scoreboard and the door policy. It could not regulate what the ranking system selects, or the objective it selects for. Sonia Livingstone of the London School of Economics, who leads the Digital Futures for Children centre, put the resulting evidentiary problem precisely in comments to the Associated Press: “Tested one by one, the evidence for each design feature will be weaker, as it is the combination.” A remedy assembled feature by feature, with the load-bearing feature excluded on constitutional grounds, is not the same intervention as a remedy directed at the system.

Nobody Is Appointed to Check the Homework

Verification will decide whether any of this means anything, and it is the part the court cut most deeply.

New Mexico asked for a court-appointed Child Safety Monitor, funded by Meta, with a term of at least five years. Biedscheid denied it, and his reasoning returns to the factory. “A monitor usually would measure pollution in the air and report whether or not identified pollutants were within satisfactory levels. But, the State asks for much more here... the monitor not only would determine which pollutants to test, but also decide what level of each pollutant would be satisfactory, and whether to force the factory ownership to use the factory to produce a new product that is more to the monitor's liking.”

In place of a monitor, the order requires Meta to file written compliance reports with the court and the state by 30 June and 31 December each year, and permits the New Mexico Department of Justice to delegate review to whatever agency it chooses. The court reserved the right to appoint a monitor or special master later if a specific requirement proves persistently contentious.

That is self-reporting with a delegated audit and a reserved power. Brooke Istook, president and chief strategy officer of the Heat Initiative, told the Associated Press that the ruling means “now, we will have someone in an official capacity checking the homework of these companies”. The order does not obviously establish who.

The wider record on independent verification of platform behaviour is worse than the order assumes. Meta shut down CrowdTangle, the analytics tool thousands of researchers and journalists used to observe content distribution on Facebook and Instagram, on 14 August 2024. Its replacement, the Meta Content Library, imposes materially tighter access conditions; a joint investigation by Proof News, the Tow Center for Digital Journalism and the Algorithmic Transparency Institute found it less transparent and less accessible, and a Coalition for Independent Technology Research survey found thirty-two of thirty-six researchers concerned the shutdown would impede their work. Europe has built the alternative: the Commission adopted its delegated act on researcher data access under Article 40 of the Digital Services Act on 2 July 2025 and opened the DSA Data Access Portal that October. New Mexico has a semiannual PDF.

Peter Ormerod, an associate professor of law at Villanova University, made the adjacent point about the age provisions: “Age verification can be done well in a privacy-protective way, but it is not easy.” Ordering a company to build a prediction model is not the same as being able to tell whether the model does what the order intended.

An Opioid Settlement for the Feed

The four hundred and twenty million dollar treatment allocation carries a framing that deserves to be stated plainly, because the court states it plainly itself. The public nuisance doctrine Biedscheid applied is the doctrine of the tobacco and opioid settlements. The remedy is abatement: you do not shut the factory, you clean up what it emitted.

The concession embedded in that framing is that the harm becomes an externality of continued production rather than a defect in the production. The order is candid about it. Paragraph 178 says accurate disclosure and education are essential “unless and until Meta eliminates the dangers to adolescents on its platforms” — a conditional assigning the elimination to Meta's discretion and the mitigation to New Mexico's clinics.

The opioid analogy also brings its own well-documented failure mode. Reporting by KFF Health News, which has spent years tracking how national opioid settlement money is actually spent, has found lax reporting rules, minimal guidance on appropriate use, and repeated instances of settlement funds substituting for existing budget obligations rather than adding to them. In Washington, DC, budget documents obtained by KFF showed around 2.3 million dollars proposed to cover the city's Medicaid contribution and at least 5.5 million to support treatment centres previously funded from general revenue. Twelve states that promised annual transparency reports produced documents that, on KFF's assessment, remain difficult for an ordinary reader to decipher.

Biedscheid anticipated some of this, noting that he expects to enter separate orders governing fund administration and drawdown. Torrez told KUNM that “elected officials” would decide how the money is spent, and that his office's focus would turn to the coming legislative session. Both are reasonable positions. Both also mean the accountability mechanism for four hundred and twenty million dollars of algorithmic harm remediation is the ordinary state appropriations process, which has not historically exerted much influence on ranking systems in Menlo Park.

The Real Number Is in Oakland, Not Santa Fe

Jury selection began on Wednesday 12 August 2026 in Oakland, before Judge Yvonne Gonzalez Rogers of the Northern District of California, in the first bellwether trial of the state attorneys general claims within MDL 3047, the consolidated federal social media adolescent addiction litigation. Twenty-nine states press the claim under the Children's Online Privacy Protection Act; four of them — California, Colorado, Kentucky and New Jersey — additionally bring claims under their own consumer protection statutes. Opening statements are scheduled for 18 August, three days from now. In late June, Gonzalez Rogers refused Meta's final attempt to dispose of the case, finding “numerous disputes of fact” over whether Meta designed its platforms to be addictive while misrepresenting what it knew, and granted the states partial summary judgment on part of the COPPA claim.

The panel now being seated will not decide the case. Gonzalez Rogers took the unusual step of empanelling an advisory jury, which will answer specific questions put to it and thereby guide a ruling she issues herself. The trial is expected to run about six weeks, with her decision expected in October. The arrangement is worth pausing on, because it reproduces the Santa Fe bottleneck in a federal courtroom a thousand miles away. In Oakland as in New Mexico, the remedial question — what, if anything, Meta must build differently — belongs to a single judge sitting in equity rather than to a jury. Whatever the advisory panel concludes about Meta's conduct, the constraint that produced paragraph 173 operates on the person who writes the order, not on the people who heard the evidence.

That is the proceeding with the capacity to change a balance sheet, and Meta has supplied the number itself. In a court filing, the company stated that the penalty calculations proposed by California, Colorado, Kentucky and New Jersey could amount to approximately 1.4 trillion dollars if those states prevail. The figure needs handling with care. It is a constructed statutory ceiling rather than a forecast of any award, arrived at by multiplying the number of affected young users by the maximum per-violation fine available under each state's consumer protection law, and legal observers do not expect anything resembling it to survive trial or appeal. What it is good for is scale. New Mexico's abatement fund is twenty hours of revenue, a sum a finance function absorbs without adjusting a single ranking parameter. A theoretical ceiling in the trillions is a different class of instrument even after two orders of magnitude have been argued off it, which is precisely what Meta will spend the next six weeks doing.

The federal MDL contains more than three thousand pending actions, claims from over two hundred and fifty school districts, and, since late 2024, more than one hundred thousand individual mass arbitration demands against Meta. The parallel California coordinated proceeding, JCCP 5255 before Judge Carolyn Kuhl in Los Angeles, encompasses roughly three thousand three hundred cases. In March 2026 a Los Angeles jury returned the first personal injury verdict in the K.G.M. case, three million dollars compensatory and three million punitive, apportioned seventy per cent to Meta and thirty to Google. The first federal school district bellwether, brought by a Kentucky district, settled in May 2026 for a combined value near twenty-seven million dollars across four companies. Three more state-court bellwethers are expected from late October.

Oakland is not the only live parallel, and it was not the first. On 27 July 2026, ten days before the Santa Fe order landed, Tennessee Attorney General Jonathan Skrmetti delivered opening arguments in the Chancery Court for Davidson County in Nashville, in the state's civil enforcement action against Meta. The claim is that Instagram was intentionally designed to promote addiction in young users, through autoplay, Reels, notifications and disappearing content, and that Meta misled the public about the platform's safety, in violation of the Tennessee Consumer Protection Act. The trial was expected to run about seven weeks. It is still running today.

What makes Nashville worth more than a passing mention is its architecture, which is New Mexico's architecture exactly. The jury decides first whether Meta violated Tennessee law. If it does, a second phase follows in which a judge, not the jury, determines both the monetary penalties and any changes Meta must make to Instagram. That is the same two-phase split that produced a 375 million dollar verdict in Santa Fe in March and paragraph 173 in August: a jury to establish that something was done, a judge to decide what may be done about it. Tennessee's remedial phase is still ahead. Should the state win the first, the second will put a Nashville chancellor in front of the question Biedscheid has just answered — whether a court of equity can order a company to rebuild a ranking system — with the same First Amendment and Section 230 arguments waiting for him, and now with a reasoned New Mexico order available as persuasive authority for the answer no.

Alongside this, the liability theory is migrating to conversational systems. Megan Garcia's wrongful death suit over the February 2024 suicide of her fourteen-year-old son Sewell Setzer III produced, in May 2025, the first ruling allowing product liability claims against a chatbot developer to proceed past the First Amendment and Section 230. Character Technologies, its founders and Google disclosed a mediated settlement of that case and four related family suits on 7 January 2026, on confidential terms and without admission of liability. Raine v. OpenAI, filed in August 2025 over the death of sixteen-year-old Adam Raine, remains pending.

Legislatures are moving on a parallel track and faster. Illinois enacted the Wellness and Oversight for Psychological Resources Act on 1 August 2025, barring the provision or advertising of therapy by anyone other than a licensed professional, with civil penalties up to ten thousand dollars per violation. Nevada's Assembly Bill 406 took effect on 1 July 2025 with penalties up to fifteen thousand. California's Senate Bill 243, effective January 2026, requires chatbot disclosure and carries a private right of action. Senate Bill 903, introduced by Senator Steve Padilla, would prohibit advertising chatbots as therapy and bar AI from making therapeutic decisions without licensed review; it passed the Senate thirty-nine to nil and cleared two Assembly policy committees before being placed on the Assembly Appropriations Committee's suspense file, where it sat against a suspense deadline of 14 August 2026 and an adjournment date of 31 August.

Torrez himself gestured at the division of labour. “The judge was actually restrained in part,” he told KUNM, “because he expects the legislature and Congress to take up some of the more sweeping calls for change that we included in this case.”

The Literature Names What the Bench Deferred

Three research strands map onto the gap the order leaves, and it is worth being precise about what each does and does not establish.

Sunil Arora, Sahil Arora and John D. Hastings submitted “The Psychological Impacts of Algorithmic and AI-Driven Social Media on Teenagers: A Call to Action” to arXiv on 19 August 2024. It is a synthesis rather than new empirical work, arguing that the algorithms underlying social platforms inadvertently produce profound psychological effects on adolescents through curated personas, notification overload and peer pressure, and calling on policymakers, developers and educators to prioritise teen safety. It is an argument for the proposition the New Mexico court accepted on the evidence, not independent proof of it.

Austin Shouli, Ankur Barthwal, Molly Campbell and Ajay Kumar Shrestha submitted “Ethical AI for Young Digital Citizens: A Call to Action on Privacy Governance” on 15 March 2025. Its framework rests on four pillars — algorithmic transparency, privacy education, ethics around parental data sharing, and accountability measures. The first and fourth are exactly what the Santa Fe order could not deliver: it produced neither a transparency mechanism capable of revealing how ranking operates nor an accountability body capable of assessing it.

The most directly relevant is “AI, Digital Platforms, and the New Systemic Risk”, submitted on 22 September 2025 by Philipp Hacker, Lilian Edwards and Atoosa Kasirzadeh, revised in May 2026 and accepted at ACM FAccT 2026. Building a framework from finance, complex systems theory, climate science and cybersecurity, the authors argue that the EU's AI Act and Digital Services Act invoke systemic risk while relying on narrow or ambiguous characterisations of it, and that current instruments miss harms such as discrimination at scale and novel failure modes arising from multiple interacting agents. Their central move — treating collective, population-level harm as a distinct regulatory object rather than an aggregation of individual injuries — is precisely the move a public nuisance claim makes, and precisely the move a court equipped only with equitable remedies against a single defendant cannot complete. Biedscheid said so himself when he reduced the fund for Meta's market share and observed that its competitors were not before him.

What Would Actually Bind an Optimiser

If a 567 million dollar abatement fund does not change the ranking system, the interesting question is what would. Four categories of instrument already exist, and none of them is a fine.

The first is deletion. Since the Everalbum consent order in January 2021, the Federal Trade Commission has required companies to destroy not only unlawfully collected data but any “affected work product” — models and algorithms trained on it. The remedy reached a children's privacy case for the first time in the 2022 action against WW International and its Kurbo subsidiary, and featured again in the 2023 Ring settlement. Algorithmic disgorgement converts a penalty into a loss of capability. It is the only American remedy that has made a model itself the object of an order, and its logic — that a firm should not retain the analytical product of unlawful conduct — maps onto engagement optimisation trained on minors without much strain.

The second is a positive duty on the recommender. The UK's Protection of Children Codes under the Online Safety Act came into force on 25 July 2025, and they do the thing Biedscheid found himself unable to do: services whose recommender systems present a medium or high risk of harmful content must configure those algorithms to exclude such content from children's feeds. That is a design mandate aimed at the ranking function, imposed by a regulator with rule-making authority rather than by a judge in a single case, and it exists because Parliament wrote it rather than because a court inferred it.

The third is a design code, and its American history is instructive about the constitutional headwind. California's Age-Appropriate Design Code Act has been in litigation since 2022. On 12 March 2026 a Ninth Circuit panel issued its second opinion in NetChoice v. Bonta, holding that NetChoice had not met the standard for facial First Amendment relief against the statute's coverage definition or its age estimation provision, while agreeing that the data-use restrictions and dark-patterns prohibition are likely unconstitutionally vague. The pattern matches Santa Fe exactly — age assurance survives, content-adjacent design rules do not.

The fourth is measurement, and it is the precondition for the other three. Article 40 of the Digital Services Act gives vetted researchers a statutory route into platform internals for the specific purpose of studying systemic risk. Nothing comparable exists in American law, which is why Biedscheid's alternative to a monitor was a report Meta writes about itself.

There is a version of this ruling that reads as a landmark, and Torrez is entitled to sell it that way. A court took an optimisation system seriously as a cause of population-level psychiatric harm, refused to let Section 230 dispose of the question, applied a doctrine built for tobacco and opioids, and attached a nine-figure number to the result. Other states and school districts now have a template, and Meta's own lawyers have to price it.

But the specific claim the ruling is being asked to bear — that ordering a company to pay for algorithmic harm changes the algorithm — is contradicted by paragraph 173 of the ruling itself. The court identified the mechanism, named the feedback loops, quoted the internal data showing the heaviest-using one per cent of New Mexico teenagers on Facebook for five hours a day, and then wrote that it granted no relief as to the design and implementation of Meta's algorithms, because to do so would be vague, competitively unfair, and probably unconstitutional. The engagement objective that produced the finding survives the finding intact, in New Mexico as everywhere else, with an accrual against it that the company had already booked.

Meaningful accountability for algorithmic harm to children would have to bind the objective function rather than the consequences of running it: a positive duty on what a feed served to a minor may optimise for, an independent capability to verify whether that duty is met, and a remedy that removes the capability rather than pricing its output. Two of those three exist in Europe. The third exists in a handful of FTC consent orders. What New Mexico has is a five-year clock that a supersedeas bond can stop, a semiannual report written by the defendant, and four hundred and twenty million dollars to treat the children the system has already reached. That is not nothing. It is a state agreeing to clean up after a process it has been told, in the same document, that it lacks the authority to alter.

References

1. State of New Mexico, ex rel. Raúl Torrez, Attorney General v. Meta Platforms, Inc., No. D-101-CV-2023-02838, “Findings of Fact, Conclusions of Law, and Judgment, Order, and Decree of the Court,” First Judicial District Court, Santa Fe County, New Mexico, 6 August 2026. https://www.courthousenews.com/wp-content/uploads/2026/08/j-bryan-biedscheid-order.pdf
2. Associated Press, “A New Mexico judge ordered new child safeguards for Meta. Advocates hope other courts follow,” The Boston Globe, 8 August 2026. https://www.bostonglobe.com/2026/08/08/nation/new-mexico-judge-ordered-new-child-safeguards-for-meta/
3. Jonathan Vandiver, “New Mexico Attorney General says Meta ruling should prompt action by lawmakers and Congress,” KUNM, 7 August 2026. https://www.kunm.org/local-news/2026-08-07/new-mexico-attorney-general-meta-ruling-lawmakers-congress
4. Meta Platforms, Inc., “Meta Reports Second Quarter 2026 Results,” StockTitan, 29 July 2026. https://www.stocktitan.net/news/META/meta-reports-second-quarter-2026-hkjfhayj8l0v.html
5. Meta Platforms, Inc., “Meta Reports Fourth Quarter and Full Year 2025 Results,” PR Newswire, 28 January 2026. https://www.prnewswire.com/news-releases/meta-reports-fourth-quarter-and-full-year-2025-results-302673127.html
6. Angela Chen, “Facebook is actually worth more thanks to news of the FTC's $5 billion fine,” MIT Technology Review, 15 July 2019. https://www.technologyreview.com/2019/07/15/134196/facebook-is-actually-richer-thanks-to-news-of-the-ftcs-5-billion-fine/
7. European Commission, “Commission finds Apple and Meta in breach of the Digital Markets Act,” ec.europa.eu, 23 April 2025. https://ec.europa.eu/commission/presscorner/detail/en/ip_25_1085
8. United States Court of Appeals for the Third Circuit, Anderson v. TikTok, Inc., No. 22-3061, 27 August 2024. https://law.justia.com/cases/federal/appellate-courts/ca3/22-3061/22-3061-2024-08-27.html
9. MDL Update, “JCCP 5255 Social Media Cases (California): K.G.M. $6M Verdict & Status August 2026,” mdlupdate.com, August 2026. https://mdlupdate.com/mdl/3047-social-media-adolescent-addiction/jccp-5255/
10. Erin Mulvaney and Isaiah Poritz, “Meta Denied Final Bid to Toss State AGs Case Before August Trial,” Bloomberg Law, 30 June 2026. https://news.bloomberglaw.com/litigation/meta-denied-final-bid-to-toss-state-ags-case-before-august-trial
11. Clare Duffy, “Character.AI and Google agree to settle lawsuits over teen mental health harms and suicides,” CNN Business, 7 January 2026. https://www.cnn.com/2026/01/07/business/character-ai-google-settle-teen-suicide-lawsuit
12. Kristen Hwang, “As AI 'therapists' dish out advice, California lawmakers try to set some limits,” CalMatters, 6 August 2026. https://calmatters.org/health/mental-health/2026/08/at-therapists-chatbot-mental-health/
13. Mintz, “Algorithmic Disgorgement: An Increasingly Important Part of the FTC's Remedial Arsenal,” mintz.com, 23 January 2024. https://www.mintz.com/insights-center/viewpoints/54731/2024-01-23-algorithmic-disgorgement-increasingly-important-part
14. Ofcom, “Tech firms must tame toxic algorithms to protect children online,” ofcom.org.uk, 24 April 2025. https://www.ofcom.org.uk/online-safety/protecting-children/tech-firms-must-tame-toxic-algorithms-to-protect-children-online
15. Cooley LLP, “NetChoice v. Bonta: Ninth Circuit Narrows Injunction Against California's Age-Appropriate Design Code Act,” cooley.com, 30 March 2026. https://www.cooley.com/news/insight/2026/2026-03-30-netchoice-v-bonta-ninth-circuit-narrows-injunction-against-californias-ageappropriate-design-code-act
16. European Commission, “Commission adopts delegated act on data access under the Digital Services Act,” digital-strategy.ec.europa.eu, 2 July 2025. https://digital-strategy.ec.europa.eu/en/news/commission-adopts-delegated-act-data-access-under-digital-services-act
17. Coalition for Independent Technology Research, “Blocking our Right to Know: Surveying the Impact of Meta's CrowdTangle Shutdown,” independenttechresearch.org, July 2024. https://independenttechresearch.org/wp-content/uploads/2024/07/CrowdTangle-Survey-Report-Final.pdf
18. Aneri Pattani and Henry Larweh, “Budget Battle Erupts in Nation's Capital Over Opioid Settlement Money,” KFF Health News, 2026. https://kffhealthnews.org/public-health/washington-dc-opioid-settlements-medicaid-budget-supplantation/
19. KFF Health News, “12 States Promised To Open the Books on Their Opioid Settlement Funds. We Checked Up on Them,” kffhealthnews.org. https://kffhealthnews.org/public-health/state-opioid-settlement-funds-transparency-update/
20. Sunil Arora, Sahil Arora and John D. Hastings, “The Psychological Impacts of Algorithmic and AI-Driven Social Media on Teenagers: A Call to Action,” arXiv:2408.10351, 19 August 2024. https://arxiv.org/abs/2408.10351
21. Austin Shouli, Ankur Barthwal, Molly Campbell and Ajay Kumar Shrestha, “Ethical AI for Young Digital Citizens: A Call to Action on Privacy Governance,” arXiv:2503.11947, 15 March 2025. https://arxiv.org/abs/2503.11947
22. Philipp Hacker, Lilian Edwards and Atoosa Kasirzadeh, “AI, Digital Platforms, and the New Systemic Risk,” arXiv:2509.17878, 22 September 2025, revised 23 May 2026. https://arxiv.org/abs/2509.17878
23. Courthouse News Service, “Jury selection begins in states' trial against Meta over youth social media harms,” courthousenews.com, 12 August 2026. https://www.courthousenews.com/jury-selection-begins-in-states-trial-against-meta-over-youth-social-media-harms/
24. JURIST, “Tennessee takes Meta to trial over Instagram addiction allegations,” jurist.org, July 2026. https://www.jurist.org/news/2026/07/tennessee-takes-meta-to-trial-over-instagram-addiction-allegations/
25. ICLG, “Meta warns of 'tens of billions' in potential exposure over algorithm-addiction claims,” iclg.com, 2026. https://iclg.com/news/23228-meta-warns-of-tens-of-billions-in-potential-exposure-over-algorithm-addiction-claims

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...