•
Plaintiffs in the social media MDL are not suing over what users posted. They are suing over infinite scroll, variable rewards, and ephemeral stories as design defects, which is the one theory Section 230 may not block.
The social media addiction cases turn on a single strategic choice. Plaintiffs are not suing Meta, Google, Snap, or TikTok over anything a user posted. They are suing over how the product is built.
That choice is not stylistic. It is the entire case, because Section 230 of the Communications Decency Act blocks claims that treat a service as the publisher of someone else's content, and it does not obviously block claims about a company's own design decisions.
This post covers the legal theory. For the current posture of the federal MDL, including the 2026 bellwether trials, see our separate write-up of MDL 3047.
A product liability claim says a thing was designed in a way that made it unreasonably dangerous, and that a safer alternative design existed. Applied here, the argument is that features like infinite scroll, autoplay, ephemeral stories, streaks, and variable-ratio reward schedules were selected to maximize time on app, that the companies knew these features had a stronger effect on adolescents whose impulse control is still developing, and that safer designs were available and rejected.
Note what the claim does not require. It does not require proving that any particular video harmed any particular child. It requires proving the design choice, the knowledge, and the causal link to a class of harms.
Section 230 says an interactive computer service may not be treated as the publisher or speaker of information provided by another content provider. Courts have read it broadly for twenty-five years.
The defense argument is that recommendation is publishing. Deciding what to show a user, in what order, is editorial arrangement of third-party content, and arrangement is exactly what the statute protects. Under that reading, calling the recommendation system a defective product is a relabeling exercise.
The plaintiffs' argument is that some of the challenged features have nothing to do with content at all. An infinite scroll works the same way whether the feed is cat videos or news. Age verification and parental controls are features of the service, not of any post. A streak counter is the platform's own creation.
Courts have generally accepted that distinction in part. Claims tied to specific content served to a specific user tend to fall. Claims about features that operate independently of content have survived. That split is why discovery in these cases focuses so heavily on internal design documents and growth metrics rather than on the feeds themselves.
Even where Section 230 does not bar a claim, the First Amendment may. Moody v. NetChoice confirmed in 2024 that a platform's curation decisions can be protected expression.
The plaintiffs' response is that a design defect claim seeks damages for the operation of a machine, not for a message. Whether that survives contact with a First Amendment defense is genuinely open, and it is the question most likely to reach the Supreme Court out of this litigation.
This is the part that makes these cases dangerous for defendants. Product liability discovery reaches internal research, A/B test results, and communications about known harms. Plaintiffs allege that internal documents show awareness of a rabbit hole effect, where a recommendation system escalates toward more extreme content because engagement rises with it.
A jury does not need to understand ranking algorithms to understand a memo. That asymmetry, between how hard the legal theory is and how simple the evidence can be, is why these cases settle.
Design defect claims against platforms have survived motions to dismiss often enough to be taken seriously, which is a change from a decade ago when Section 230 ended these cases at the pleading stage. They have not yet produced a defense verdict or a plaintiff verdict at trial in the federal MDL, because the cases set for trial in 2026 settled first.
That is worth sitting with. A theory that never reaches a jury never gets tested, and settlements produce money without producing law.
The federal cases are consolidated in the Northern District of California. State court actions run in parallel and are not on PACER.
Pull the federal filings here: social media adolescent addiction litigation. MDL dockets generate hundreds of entries a month, most of them routine, so case alerts filtered to a case are far more practical than reading the docket.
Because so much of an MDL turns on how one judge manages it, the assigned judge's record on Daubert motions and bellwether selection is worth pulling from the judge lookup before you predict anything.
Design defect claims require plaintiffs to identify a safer alternative that was available and feasible. Vague assertions that a product should have been less addictive do not satisfy that element.
The alternatives plaintiffs point to are concrete: chronological feeds instead of engagement-ranked ones, default time limits for accounts identified as belonging to minors, age verification at signup rather than a self-reported birthdate, removal of streak mechanics and autoplay for minor accounts, and notification batching instead of continuous interruption.
Several of these exist as optional settings on the platforms already, which is the point plaintiffs press hardest. A feature the company built and buried in a settings menu is difficult to describe as infeasible.
Personal injury claims resist class treatment because causation and damages differ for every plaintiff. Two teenagers with the same diagnosis have different histories, and no common proof resolves both claims.
So the cases proceed as a multidistrict litigation: consolidated for pretrial purposes before one judge, then returned to their home districts for trial if they do not settle. Bellwether trials substitute for the class device by producing information about value.
School district and state attorney general claims are different animals again. Those plaintiffs assert their own injuries, public nuisance and expenditure of public resources, which do not require individualized proof of harm to any particular child.
The state enforcement version of the same theory is covered in State of Texas v. TikTok. For another dispute over whether software output is the vendor's own conduct or someone else's, read Amazon v. Perplexity. And for the Supreme Court's 2026 word on when an intermediary answers for what its users do, see Cox Communications v. Sony Music.