What the Outrage Study Shows About Algorithm Liability

Last verified: August 26, 2026. This is legal-background analysis, not legal advice. The useful starting point is a negative one: the 2021 outrage-amplification paper helps explain a plausible mechanism by which engagement feedback can change later expression, but no located court order cites Brady et al. by name as the basis for the 2026 liability outcomes reviewed here.

That distinction matters because the public shorthand has become too loose. A study showing that outrage-specific feedback predicts more later outrage is not the same thing as trial proof that Meta or Google caused a particular injury, knew a particular product design was dangerous, or lost a particular Section 230 argument. The 2026 litigation materials point elsewhere: to platform design, engagement-maximizing architecture, internal research, whistleblower evidence, and the procedural posture of each case.

What the outrage paper can fairly show

Brady, McLoughlin, Doan, and Crockett studied social learning and moral-outrage expression in online social networks using 7,331 Twitter users and 12.7 million tweets. The paper reported two preregistered observational studies and two experiments, with an estimated 2% to 3% expected increase in next-day outrage expression for every 100% increase in outrage-specific feedback on the previous day. [1]

That is a concrete mechanism, not a verdict form. The study gives lawyers a more disciplined way to describe one pathway that otherwise gets flattened into “the algorithm made people angrier.” Engagement feedback, in the study’s account, can act as a social-learning signal: when users receive more feedback for outrage, later outrage expression becomes more likely. That proposition can help with plausibility, expert framing, discovery relevance, and corporate-knowledge questions.

The authors’ own limitations are not footnote housekeeping; they are the guardrails for litigation use. The observational studies could not draw causal inferences, Twitter users were not representative of the general population, and the dictionary-of-condemnation classifier was about 75% accurate on a training set of roughly 26,000 tweets and not generalizable without additional validity testing. [1]

A brief that uses Brady et al. carefully can say the paper supports a mechanism by which engagement feedback may reinforce outrage expression at network scale. A brief that says the paper “proves” platform liability in the 2026 cases is doing more work than the paper, or the located court record, will bear.

The litigation target is product design, not the mere presence of harmful posts

The important legal movement in 2026 is not that courts suddenly treated all harmful social media content as actionable. The sharper move is that courts and juries were asked to treat engagement-maximizing design as the product feature at issue: recommendation systems, feedback loops, default settings, notifications, metrics, and other architecture that allegedly encourages compulsive use or routes users toward harmful experiences.

That is the hinge for Section 230. Claims aimed at a platform’s role as publisher or speaker of third-party content remain exposed to the familiar immunity argument. Claims aimed at the platform’s own design choices ask a different question: whether the defendant is being sued for user speech, or for building and operating a product architecture alleged to cause injury.

The corporate-evidence trail is where that distinction becomes operational. A BBC whistleblower investigation reported internal Meta research stating that “our algorithms presume that users like that content and want more of it,” and reported internal figures showing Reels hostile-comment prevalence 75% higher for bullying and harassment, 19% higher for hate speech, and 7% higher for violence and incitement. The same reporting quoted an internal assessment that “the current set of financial incentives our algorithms create does not appear to be aligned with our mission,” and discussed Zuckerberg’s February 18, 2026 testimony and “Project Myst.” [2]

Those materials do different work from the Brady paper. They are not just general social-science background. They are closer to notice, design knowledge, internal risk evaluation, and the reasonableness of continuing or modifying a design after a company has information about how the system behaves.

The FTC’s September 2024 staff report supplies related governance context. FTC staff reported that nine large social media and video-streaming companies fed user and non-user data into algorithms with little or no opt-out and had inadequate monitoring and testing practices. [3] That report is not a substitute for causation proof in any individual injury case. It is useful because it places algorithmic data use, testing, monitoring, and opt-out design into an institutional record before the 2026 verdict wave.

Verified litigation status map as of August 26, 2026

KGM is a verdict, not an appellate endpoint

KGM is the case most likely to be overread in a client alert because it has the cleanest headline. NPR reported that a Los Angeles jury returned a March 25, 2026 verdict against Meta and Google in a social-media-addiction case, awarding $3 million in compensatory damages and $3 million in punitive damages, with fault allocated 70% to Meta and 30% to Google. [4]

Post-verdict coverage reported that the judge refused to overturn the $6 million verdict. [5] The Conversation described the case as a landmark social-media-addiction loss for Meta and Google and discussed the expected fallout, including appeals. [6] That is enough to treat KGM as a major jury event. It is not enough to write as though an appellate court has endorsed every theory in the case.

The safest procedural label is therefore: first reported defective-product-style jury verdict against major social media platforms, post-trial challenge rejected, appeals pending or announced, no located appellate merits ruling yet. The liability implications are substantial because a jury accepted a product-design framing. The doctrinal implications remain provisional until appellate treatment arrives.

New Mexico carries public-nuisance and consumer-protection weight, with a source caveat

New Mexico v. Meta is the harder case to summarize because the numbers are larger and the source discipline is more important. The Guardian reported on March 24, 2026 that a New Mexico jury found Meta liable in the state’s case. [8] The New Mexico Department of Justice later announced that the court ordered Meta to pay $942 million and overhaul protections for children on Facebook and Instagram, describing 75,000 Unfair Practices Act violations, $375 million in civil penalties, a $567 million abatement fund, and five years of court-supervised design reforms. [7]

That should be treated as a major public-nuisance and consumer-protection outcome. It should also be cited with the correct source label: the figures here come from the state DOJ release and corroborating press coverage, not from a crawled copy of the judgment order. The reported final-judgment timing in early August 2026, including August 6–7 references in coverage, should be re-checked against the judgment itself before being used in a filing.

The New Mexico result matters because public nuisance and unfair-practices theories do not require the same proof architecture as an individual design-defect claim. They can focus on statewide conduct, youth-protection representations, design practices, remediation, and abatement. But they still benefit from the same design-content split: the actionable conduct is framed around Meta’s systems and business practices, not merely the existence of harmful third-party posts.

Massachusetts and the Ninth Circuit show why Section 230 is not doing all the work it once did

The Massachusetts Supreme Judicial Court ruling is the cleanest cited example of the design-feature point. In Commonwealth v. Meta, reported April 10, 2026, the court held that Section 230(c)(1) did not immunize claims based on Instagram’s design features alleged to injure children. [9]

The Ninth Circuit development points in the same procedural direction, though with different posture. Daily Journal coverage reported that, on August 10, 2026, the Ninth Circuit rejected an immediate Section 230 appeal, allowing claims against Meta to proceed toward the four-state bellwether before Judge Yvonne Gonzalez Rogers. [10] NPR then reported that the trial involving California, Colorado, Kentucky, and New Jersey opened in August 2026, with the states seeking up to $1.4 trillion. [11]

For briefing purposes, these rulings should not be collapsed into a single sentence saying “Section 230 no longer applies.” The more defensible formulation is narrower: courts are allowing certain claims to proceed where the alleged wrong is platform design or business conduct rather than the publication of third-party content. That formulation leaves room for claim-specific analysis, jurisdictional differences, appeal posture, and the possibility that some theories may still fail on the merits.

For related state-law comparisons on youth-protection and platform-design obligations, see Do Instagram Teen Account Settings Comply with State Privacy Laws?. The First Amendment cases involving school social-media restrictions are a different line of authority, but they are useful reminders that content, platform design, and government regulation of expression cannot be treated as interchangeable categories.

How the main claim types use the same design-content distinction differently

This is where the outrage research remains useful but bounded. The Brady paper can help explain why engagement feedback might plausibly reinforce later expression. Internal documents can show what a company studied, knew, discounted, or chose not to change. Court orders, verdicts, and appellate rulings determine whether those materials actually supported liability in a particular procedural posture.

What not to say in a motion, memo, or media quote

The weakest sentence in this area is also the most tempting: “The outrage study caused the courts to hold Meta and Google liable.” It confuses mechanism evidence with operative proof and then confuses reported verdicts with settled doctrine.

A more accurate formulation would be: peer-reviewed outrage-amplification research supports the plausibility that engagement feedback can shape later user expression; the 2026 liability materials, as located, turned on platform-design theories, internal research, whistleblower evidence, consumer-protection or nuisance frameworks, and Section 230 rulings that distinguished design features from publication of user content.

That phrasing preserves three necessary separations. Mechanism evidence is not operative proof. Corporate knowledge is not identical to causation. A jury verdict, a state-court judgment described by an attorney general, an appellate immunity ruling, and an ongoing bellwether trial do not carry the same procedural weight.

The briefing line that survives the record

As of August 26, 2026, the actionable target in the reviewed materials is engagement-maximizing architecture and platform design, not the mere presence of harmful user posts. That is the distinction allowing design-defect, public-nuisance, and consumer-protection claims to move past Section 230 arguments in the cases and rulings discussed above.

But the Brady outrage study should not be cited as though it were the operative trial evidence in those outcomes. Use it for mechanism and plausibility. Use internal research, whistleblower materials, executive testimony, and design documents for notice and knowledge. Use court orders, verdict records, and appellate decisions for procedural reality.

References

1. How social learning amplifies moral outrage expression in online social networks — Science Advances / PMC, 2021.
2. Meta accused of censoring evidence of harm to children in virtual reality — BBC, March 16, 2026.
3. FTC Staff Report Finds Large Social Media and Video Streaming Companies Have Engaged in Vast Surveillance of Users with Lax Privacy Controls and Inadequate Safeguards for Kids and Teens — Federal Trade Commission, September 2024.
4. Meta, YouTube face $6 million verdict in landmark social media addiction case — NPR, March 25, 2026.
5. Judge Upholds $6 Million Social Media Verdict Against Meta, Google — Beasley Allen.
6. Meta and Google just lost a landmark social media addiction case. A tech law expert explains the fallout — The Conversation.
7. Court Orders Meta to Pay $942 Million and Overhaul Protections for Children on Facebook and Instagram in Landmark New Mexico Ruling — New Mexico Department of Justice.
8. Meta found liable in New Mexico lawsuit over child safety on Facebook and Instagram — The Guardian, March 24, 2026.
9. In Massachusetts, Section 230 Does Not Immunize Meta from Claims that Instagram’s Design Features Injure Children — Crowell & Moring, April 10, 2026.
10. 9th Circuit rejects Meta, TikTok bid for immediate appeal of Section 230 rulings — Daily Journal, August 10, 2026.
11. Meta trial over kids’ social media addiction could cost the company up to $1.4 trillion — NPR, August 17, 2026.