Background
The plaintiffs in this putative class action are retail investors who were targeted on Facebook and Instagram with advertisements for “investment groups” supposedly run by recognizable Wall Street figures — including impersonated likenesses of Kevin O’Leary and Bank of America’s Savita Subramanian — promising daily returns of 30 to 40 percent. Clicking the ads sent users into private WhatsApp groups (also a Meta product), where scammers posing as financial advisers persuaded them to buy shares of China Liberal Education Holdings Ltd. (ticker CLEU). On January 22, 2025, the scammers began pumping CLEU shares from $5.32; by January 29 the price had risen above $7.90; on January 30 the market learned of a 240-million-share secret offering held by the scammers’ co-conspirators, and CLEU collapsed to $0.15. Class losses are alleged to exceed $300 million.
Plaintiffs sued Meta on behalf of the class, asserting aiding and abetting fraud, negligence, breach of contract, an Unruh Act civil-rights claim, unjust enrichment, and alternative claims for promissory estoppel and breach of the covenant of good faith and fair dealing. Meta moved to dismiss on Section 230 grounds and, alternatively, on the merits.
The Court’s Holding
Chief Judge Seeborg denied the motion in part and granted it in part with leave to amend. The Section 230 ruling, and its specific reliance on Meta’s generative-AI ad-creation tools, is the most novel and consequential piece of the order.
Section 230: Meta’s AI tools raise a factual dispute on material contribution
Following the Ninth Circuit framework from Fair Housing Council v. Roommates.Com, the court examined whether Meta was a co-developer of the offending advertisements. Roommates.Com tells courts to find that a website “helps to develop unlawful content” — and so loses Section 230 immunity — only when it “contributes materially” to the content’s alleged illegality.
The plaintiffs pleaded that Meta’s Ads Manager suite includes three tools that did exactly that:
- Flexible Format — Meta automatically optimizes the ad and selects which images, layouts, and platforms (Facebook or Instagram) to show.
- Dynamic Creative — takes multiple advertiser-supplied images, videos, text components, and calls-to-action and mixes and matches them into personalized creative variations.
- Advantage+ Creative — uses generative AI to add “creative enhancements” including AI-generated text and images, applying overlays, modifying image backgrounds, generating ad copy variations, and inserting call-to-action buttons (e.g., a link to join a WhatsApp group).
The court found that the third tool was decisive at the pleadings stage. The complaint alleged that the CLEU scammers used Advantage+ Creative to deploy at least 86 different variations of ads featuring the Subramanian image, and described a separate experiment in which a journalist fed Advantage+ Creative a prompt about “10% weekly returns” and the tool generated additional fraudulent ad copy on its own, including new sentences like “Tired of living paycheck to paycheck? Break the cycle and start earning steady weekly income with our proven system.” The court emphasized: the reporter did not write that language; Meta did.
Distinguishing Carafano v. Metrosplash — where Section 230 protected an online dating site that offered structured questionnaire responses because “the selection of the content was left exclusively to the user” — the court explained that here, by contrast, plaintiffs averred that Meta itself generated some of the false statements. That is precisely the situation in which, even Carafano’s dictum acknowledged, Section 230 would not apply. Following the recent district court decision in Forrest v. Meta Platforms (which held similarly on comparable facts), the court concluded that Meta’s averred role in “literally generating, using artificial intelligence, the images and text in the advertisements” is not protected by Section 230. Section 230 therefore did not bar the case at the pleadings stage.
Aiding and abetting fraud and negligence: claims survive
The court held that plaintiffs plausibly pleaded that Meta had “actual knowledge” of the scammers’ conduct and substantially assisted it. Two strands supported that holding. First, Meta’s automated ad-review system was alleged to scan each ad’s components for policy violations; an ad featuring a Bank of America executive promoting daily returns three to four times the average annual U.S. equity return is “facially ridiculous” and would have flagged real knowledge of the fraud. Second, the act of generating the fraudulent text through Advantage+ Creative was itself alleged to have given Meta knowledge of what it was helping to create. The court also rejected Meta’s nonfeasance/economic-loss-rule arguments on the negligence claim, holding that the plaintiffs adequately pleaded misfeasance: Meta “actively assisted in the creation of the fraudulent advertisements,” making the position of victims worse.
Breach of contract and Unruh Act: dismissed
The court dismissed the breach-of-contract claim because Meta’s Terms of Service provisions cited by plaintiffs — rules prohibiting scam content and aspirational language about Meta’s efforts to combat fraud — do not create a binding contractual obligation on Meta to police its platform. Promissory-estoppel and good-faith-and-fair-dealing alternatives failed for the same reason: they cannot manufacture a contractual promise that the ToS itself does not contain.
The court also dismissed the Unruh Act civil-rights claim. Plaintiffs alleged that Meta’s ad tools targeted them because of their race or national origin (the scam ads featured Subramanian and similar figures), but the Unruh Act prohibits exclusion based on protected characteristics, not inclusion. The plaintiffs were targeted, not denied access.
Unjust enrichment
The unjust-enrichment claim survived because it sounds in quasi-contract restitution and is properly pleaded in the alternative to the (failed) breach-of-contract claim under Federal Rule of Civil Procedure 8(d)(2).
Key Takeaways
- Generative-AI advertising tools may break Section 230 immunity. This may be the first published district-court order to point squarely at AI-generated ad copy — not just algorithmic targeting or matching — as the basis for the “information content provider” carve-out. Platforms whose ad systems produce text, imagery, or call-to-action language on their own should expect the same pleading-stage analysis when their tools are alleged to have contributed to fraudulent content.
- The factual dispute the court identified is narrow but pointed. The court did not hold that Meta is a publisher of the ads, that algorithmic targeting alone defeats Section 230, or that ad-review failures create liability. It held only that plaintiffs plausibly alleged Meta itself generated some of the offending text and that this allegation, if borne out, would suffice under Roommates.Com to remove Section 230’s shield as to those specific ads.
- The aiding-and-abetting-fraud theory has new traction in advertising-fraud cases. The court permitted “actual knowledge” to be inferred at the pleadings stage from (i) the facial absurdity of an ad’s claims and (ii) Meta’s averred role in generating the ad language. Defendants relying on automated review systems should expect to face questions about whether those systems amount to deliberate ignorance.
- The decision broadens the practical effect of an existing N.D. Cal. line (Forrest v. Meta, Calise v. Meta) and aligns with the parallel sister case Suddeth v. Meta Platforms (also denying Meta’s Section 230 motion in a related CLEU pump-and-dump). Meta is now defending multiple class actions involving generative-AI ad tools and may seek interlocutory appeal of Section 230 immunity once a controlling question of law crystallizes.
Why It Matters
Generative-AI features have spread across nearly every advertising platform in the last two years. Platforms compete on how aggressively their AI can transform an advertiser’s input into multiple, personalized creative variations. This opinion tells those platforms that the more their AI produces actual content — words, images, calls to action — the harder it becomes to characterize the resulting ad as content “provided by another information content provider” under Section 230(c)(1). The same logic that has long kept user-generated content shielded does not automatically extend to ad copy that the platform’s own model wrote.
For platform legal teams, the practical implication is to audit whether AI ad-generation features actually produce content that crosses the “material contribution” line under Roommates.Com, especially when scammers are alleged to have used those features to spin up large-scale fraudulent campaigns. For plaintiffs’ counsel in similar cases, the order is a near-template for the §230 analysis to be pleaded around generative-AI tools.
Source
The court’s published order is available here: Bouck v. Meta Platforms, Inc. — Order Denying in Part and Granting in Part Motion to Dismiss Second Amended Complaint (March 24, 2026).