Meta's Social-Media-Addiction Settlement

Overview

On August 26, 2026, Meta agreed to a settlement — reported in headlines as "up to $18 billion" — with a coalition of state attorneys general, resolving the federal jury trial that had been underway since August 18 in the U.S. District Court for the Northern District of California (Oakland).¹ The case is People of the State of California v. Meta Platforms, Inc., No. 4:23-cv-05448-YGR, consolidated with the related MDL, In re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, No. 4:22-md-03047-YGR — both before Judge Yvonne Gonzalez Rogers.² Meta did not admit liability or wrongdoing.³

A note on sourcing. This page is built primarily from the actual court filings — the Joint Motion to Enter Consent Judgment (Docket #572) and its attached Settlement Agreement (Docket #572-1, 130 pages), both filed Aug. 26, 2026, plus the New Mexico state-court judgment and the New Mexico DOJ's press release — rather than from news paraphrase alone. Where a claim rests only on a secondary news source, that's noted. See the Open Questions section below for what still needs primary-source verification. The two central figures below — the $942M New Mexico judgment and the court's rejection of Meta's Section 230 defense — were independently confirmed against the judgment PDF while drafting this page.

What the States Alleged

• Meta designed Facebook and Instagram with features — infinite scroll, algorithmic recommendation, frequent notifications — that "entice, engage, and ultimately ensnare youth and teens," and that the company "repeatedly misled the public about the dangers of its Platforms."⁴
• Meta violated the Children's Online Privacy Protection Act (COPPA) and state Unfair and Deceptive Acts and Practices (UDAP) statutes by collecting, retaining, and using data from users under 13.⁴
• Meta disputed all of this and "asserted that its actions were reasonable and lawful at all times."⁴

Who Actually Sued, and Who Actually Settled — Two Different Numbers

News coverage has cited both "29 states" and "52 attorneys general," and both are correct — they're just describing different things. (Without reading the Settlement Agreement's defined terms, the two figures look like a contradiction. They aren't: the settlement folds in additional states that had separate, parallel litigation — the Satellite AG Actions — plus any other state that chose to join the Agreement without ever filing suit here.)

• 29 states were the actual litigating plaintiffs who filed the Oct. 24, 2023 complaint and signed the Aug. 26, 2026 Joint Motion and Consent Judgment in this case: Arizona, California, Colorado, Connecticut, Delaware, Hawaiʻi, Idaho, Illinois, Indiana, Kansas, Kentucky, Louisiana, Maine, Maryland, Minnesota, Nebraska, New Jersey, New York, North Carolina, Ohio, Oregon, Pennsylvania, Rhode Island, South Carolina, South Dakota, Virginia, Washington, West Virginia, and Wisconsin.⁵
• The Settlement Agreement itself defines a much larger universe of up to 51 eligible jurisdictions (essentially all states, D.C., Puerto Rico, American Samoa, and the Northern Mariana Islands, except Florida and New Mexico) who can join as "Settling States" — including states that never filed suit in this case at all, via 13 separate "Satellite AG Actions" already pending in state courts (Arkansas, D.C., Iowa, Massachusetts, Mississippi, Montana, Nevada, New Hampshire, Oklahoma, Puerto Rico, Tennessee, Utah, and Vermont) that the same Agreement also resolves.⁶ The press's "52 AGs" figure is almost certainly this broader roster, not the 29-state MDL plaintiff list.
• The Agreement sets a "Minimum State Participation" threshold: at least 40 states or territories had to actually sign on for the deal to take effect at all.⁷

Florida and New Mexico: Not Just Non-Signatories, Structurally Excluded

Florida and New Mexico do not appear anywhere in the Settlement Agreement's defined list of eligible "Settling States" — meaning they weren't offered a seat at this particular table, not merely declining one.⁶ New Mexico's absence has a clear reason: it had already litigated its own case to judgment.

New Mexico's parallel case, State of New Mexico ex rel. Torrez v. Meta Platforms, Inc., No. D-101-CV-2023-02838 (1st Jud. Dist. Ct., Santa Fe County), produced a larger and more punitive outcome than this settlement:

• Phase 1 (jury trial, Feb. 2 – Mar. 24, 2026): the jury found Meta committed 75,000 violations of New Mexico's Unfair Practices Act and imposed the maximum $5,000 penalty per violation — $375,000,000 — formalized in an "Order Correcting Verdict Form" entered March 26, 2026.⁸
• Phase 2 (bench trial, May 4–22, 2026): the court additionally found Meta's platforms constitute a public nuisance in New Mexico and — notably — rejected Meta's Section 230 defense, ordering $567,000,000 in further abatement funding.⁹
• The final judgment combining both phases ($942 million total) was entered August 6, 2026, 4:41 PM, by Judge Bryan Biedscheid.¹⁰
• The court's injunctive relief runs for five years and includes overnight notification blackouts, hidden "like" counts, mandatory time limits, and enhanced sextortion/exploitation protections for New Mexico minors — broadly similar in shape to the nationwide settlement's terms, described below.¹¹
• Notably, the court explicitly declined to mandate hard age-verification, ruling that "sweeping policy choices of that kind belong to the legislative and executive branches, not the courts" — AG Raúl Torrez has said he'll pursue that gap through New Mexico legislation instead.¹¹ This is a meaningful limitation worth flagging in any framing of the NM case as a total win: the court itself said the remedy is incomplete without a legislative fix.

Because New Mexico already secured a stronger, court-ordered outcome — including the Section 230 ruling Meta avoided nationally — joining the multistate settlement would have meant trading a bigger win for a smaller, negotiated one, which is presumably why it wasn't offered a seat, and why Torrez was openly critical of the nationwide deal (see Key Quotes below). Florida's absence has no equivalent documented explanation here.

Financial Terms

The headline figure varies by outlet ($16.7B–$18B) because it's a 10-year, partly contingent structure, not a single check:

The contingent portion unlocks only once "Industry-Wide Adoption" occurs — defined in the Agreement as Snap, TikTok, and YouTube all becoming subject to substantively equivalent time-management and age-assurance obligations, whether by their own settlements, new laws, or voluntary compliance verified by an independent auditor.¹⁴ If that happens, Meta's own restrictions tighten further (see Product-Change Requirements, below) — it's a second, stricter phase of Meta's own obligations that activates on industry adoption, not a one-time drop from a two-hour cap to a one-hour cap.¹⁴

A separate, smaller settlement is bundled into the same document. The Agreement also resolves four pre-existing "Cambridge Complaints" tied to the 2016 Cambridge Analytica data-sharing scandal — filed by California (Dec. 18, 2025), New Mexico (Jan. 21, 2021), the District of Columbia (Dec. 19, 2018), and Illinois (amended Oct. 3, 2019) — for a defined "Cambridge Settlement Amount" of exactly $459,293,017.80, paid to a separately defined list of "Cambridge Settling States."¹⁵ This is not part of the $17–18B headline figure, and its inclusion means New Mexico is a beneficiary of part of this settlement even though it isn't a "Settling State" for the addiction claims themselves.

Product-Change Requirements

The Settlement Agreement's injunctive terms are phased: Phase I applies immediately (from the Compliance Date, six months after the Effective Date); Phase II applies only once Industry-Wide Adoption is triggered.¹⁶ A single number for each term, as most news coverage gives, erases that structure.

• Age Assurance: Meta must deploy an age-assurance framework within one year, but the required accuracy is itself phased and imperfect — the contractual ceiling on the false positive rate (share of actual teens wrongly cleared as adults) starts at 14% for 16–17-year-olds / 7% for 13–15-year-olds in Year 1, tightening to 10% / 5% by Year 2.¹⁷ Even at target, roughly 1 in 10 older teens could still be misclassified as an adult.
• Night Access Mode: Phase I — full app blackout 12:00 AM–6:00 AM, plus push notifications disabled separately from 10 PM–7 AM; Phase II (post industry adoption) — the full blackout itself expands to 10 PM–7 AM.¹⁸
• Daily Time Limit: Phase I — 2 hours/day cumulative across Facebook + Instagram, excluding messaging, settings, and "Longform Content" (22+ minute videos); Phase II — 60 minutes per app, 120 minutes cumulative (tighter, not simply half of Phase I).¹⁹
• School Mode: push notifications disabled 8 AM–3 PM, Monday–Friday, Aug. 15–June 15 ("School Hours" as defined), adjustable by a parent.²⁰
• Feed Options: teens must be given a "Reasonably Accessible" option to switch to a non-personalized, chronological feed — an available choice Meta must surface, not a new default applied automatically.²¹
• Social Comparison: like/reaction counts hidden from teens by default, changeable only with parental approval.²²
• "Cosmetic Procedure Filter" ban: teens are blocked from using filters that "distort, sculpt, redefine, or idealize a user's face in a way that cannot be achieved without cosmetic surgery or extreme makeup techniques" — a precisely defined term that explicitly excludes ordinary makeup/smoothing filters and fantasy/avatar effects.²³
• Independent Auditor: a third party, jointly selected by Meta and a bipartisan "State Committee," audits compliance for the 10-year term.²⁴
• No private right of action: the Agreement explicitly creates no new right for individual families or non-party states to sue over its terms — enforcement is reserved to the signing AGs and the court.²⁵
• Meta is enjoined from making "false, misleading, or deceptive representations" about its safety features going forward — but the release only covers representations made on or after the Effective Date, not past ones.²⁶

How the Settlement Landed

The federal docket shows the jury trial ran August 18–26, 2026 (with a scheduled Aug. 20 session vacated and moved to Aug. 24), and that on August 26 — the same day — the court entered a "Jury Trial – Completed" minute entry (49 minutes in court, unusually short compared to the 5+ hour trial days preceding it), followed immediately by: a "Trial Order No. 3 Re Trial Proceedings, Timing of Juror Discharge, & Privilege Dispute," the Joint Motion to Enter Consent Judgment, the entered Consent Judgment itself, and case termination — with a "Jury Note" logged the next day.²⁷

That sequencing — a juror-discharge order and a jury note appearing at the exact moment of settlement — is suggestive that this deal was struck while the jury was still deliberating, possibly mid-note to the court, rather than settled well in advance of a verdict. This is inference from docket sequencing, not a confirmed fact — see Open Questions.

Key Quotes

"Meta has agreed to make massive transformations that will reduce the risk of harm from its platforms — and will do it within months." — Rob Bonta, California Attorney General²⁸

"The payouts are peanuts compared to the profound harms Meta's profit-driven addictive features inflicted on kids. We'll see them at trial." — James Uthmeier, Florida Attorney General (state not part of the settlement's eligible-jurisdiction list at all)²⁹

"We had hoped a nationwide settlement might echo the full strength of the protections New Mexico secured in court — including a direct ban on romantic and sexualized AI chatbot interactions with minors." — Raúl Torrez, New Mexico Attorney General³⁰

"Meta built products it knew would fuel addiction, deepen a youth mental health crisis, and expose children to sexual exploitation, then lied to parents and policymakers about the danger. Today, it pays for that choice." — Raúl Torrez, on the separate New Mexico judgment¹¹

Context and Significance

• The settlement averted Mark Zuckerberg's expected testimony; Adam Mosseri (head of Instagram) had already testified, per secondary reporting, on August 25 — a date consistent with a federal trial day the docket confirms occurred that day, though the docket entry itself is a generic minute entry and doesn't name witnesses.³¹
• The document itself calls this the states' full resolution of their COPPA and UDAP claims against Meta in this action — not, contrary to some framing, a broader industry-wide product-safety mandate; the contingent terms bind only Snap, TikTok, and YouTube by name ("Core Industry Members"), and only if they separately agree or become legally obligated.¹⁴
• This is a separate matter from the 2020 antitrust lawsuit brought by a coalition of state AGs (commonly cited as ~48 states) over Facebook's acquisitions of Instagram and WhatsApp — worth stating plainly since the two cases are easy to conflate by AG-count alone.
• This case is squarely on this site's thesis: New Mexico's court found that Section 230 does not shield Meta from liability for a public nuisance caused by its recommender-driven product design — precisely the argument developed at /section-230 and /recommender-systems.

Open Questions

1. Confirm the "jury deliberating at settlement" inference by pulling Docket #575 (Trial Order No. 3, re timing of juror discharge) and #578 (Jury Note) directly — both exist on CourtListener but weren't downloaded for this draft.
2. Cite the entered judgment, not the proposed one. The Settlement Agreement text used throughout this page comes from Docket #572-1, filed as the parties' "[PROPOSED] Consent Judgment." The court's actual signed order is a separate docket entry, #576, not yet pulled. It should be functionally identical, but hasn't been diffed against the proposed version.
3. Verify the full 51-jurisdiction defined list against an actual final signatory count. This page counted the Settlement Agreement's defined universe by hand from a definitions section; an official state-by-state participation list (if AG offices publish one) would let us state the true settling-state count with certainty.
4. Florida's exclusion has no documented reason in the sources reviewed here — unlike New Mexico's (a stronger parallel judgment), nothing in the Settlement Agreement or news coverage gathered explains why Florida isn't in the eligible-jurisdiction list.
5. The Oct. 15, 2024 motion-to-dismiss ruling is currently sourced only to a secondary Findlaw link; see the companion timeline page for detail on the cross-appeals this raises.
6. The MDL docket itself (No. 4:22-md-03047-YGR, motion at Docket #3447) hasn't been reviewed — it likely contains the full picture of the 13 Satellite AG Actions and any individual/school-district personal-injury claims consolidated alongside the state AG claims.
7. Confirm the June 2026 summary-judgment date precisely. The only source located for this (a California AG press release) is actually about an unrelated Aug. 10, 2026 Ninth Circuit appeal dismissal and only mentions the June order in passing.

See Also

For the full docket-by-docket timeline, the complete 29-state plaintiff list, and notes on what's still unverified, see Timeline: State AGs v. Meta.

Note on the dollar figure: outlets reported anywhere from $16.7B (CNBC) to $18B (Meta's own framing); this page treats "$17.1–18B" as the reasonable range rather than picking one figure, consistent with the Financial Terms table above, which is sourced to the Agreement's own defined payment structure rather than to any single outlet's rounding.

References