Meta's AI Ads: What the Fraud Rulings Actually Decided

An AI advertising tool can change more than the speed of a campaign. If it helps create the misleading part of an advertisement, that involvement can matter to liability. But the decisions in Bouck v. Meta Platforms do not establish that an AI generator automatically becomes the legal author of a fraudulent statement.

Corrected September 8, 2026: this article now distinguishes the March pleading decision from the June dismissal and removes the earlier claim that federal securities fraud liability had been established.

What Judge Seeborg Actually Ruled on March 24

In the March 24, 2026 order in Bouck (No. 25-cv-05194-RS), Chief Judge Richard Seeborg considered allegations that fraudulent advertisements led investors into WhatsApp groups and ultimately into purchases of China Liberal Education Holdings shares. The plaintiffs alleged that Meta’s advertising tools helped create the misleading text and images.

The court allowed state-law claims for aiding and abetting fraud, negligence and unjust enrichment to proceed at that stage. It found deficiencies in the contract and California Unruh Act claims and granted leave to amend. These were allegations evaluated for pleading sufficiency; the judge did not find that Meta had committed fraud.

There was a significant later ruling. On June 11, 2026, the court dismissed the case without prejudice under the Securities Litigation Uniform Standards Act (SLUSA). The alleged misrepresentations were sufficiently connected to purchases of covered securities for SLUSA to bar the state-law class action in that form. The order allowed refiling under federal securities law or in another form not precluded by SLUSA. It did not decide that such a new claim would succeed.

How Section 230 Applies to Content Generation

The March order applied the distinction between publishing someone else’s information and materially contributing to the alleged illegality of that information. The complaint described more than selling ad space: it alleged that Meta’s generative tools produced some of the false statements and images themselves.

Those pleaded facts created a dispute about whether Section 230 protected Meta. The court’s reasoning depended on the alleged contribution to the unlawful content. It should not be compressed into a universal rule that all targeting is immune or that any use of a generative tool removes immunity. The tool’s role, the particular information and the cause of action still matter. The June SLUSA dismissal also shows why defeating one defence does not establish liability.

The Rule 10b-5 Maker Problem

A federal securities claim raises separate questions. In Janus Capital Group v. First Derivative Traders, 564 U.S. 135, 142–143 (2011), the Supreme Court tied the maker of a statement, for Rule 10b-5(b), to ultimate authority over its content and communication. Preparing language for someone who retains that authority is not enough by itself.

An AI system generating text therefore does not automatically make its vendor the statement’s legal maker. The relevant inquiry concerns the people or entities exercising authority over the statement. Nor does identifying a maker dispense with the other requirements of the claim. The Bouck orders discussed here did not apply Janus to hold Meta liable under Rule 10b-5. Treating that possible future theory as an existing holding would get the case backwards.

What This Means for a Regulated Advertiser

For a firm using AI to draft advertisements, my practical inference is to keep responsibility for a claim identifiable. Before publication, someone should verify the evidence behind a performance figure, the conditions attached to a guarantee and the accuracy of a testimonial. The applicable duties depend on the firm’s jurisdiction, regulated status and communication.

For example, if a model turns an approved historical return into a promise of future returns, the review needs to catch the change in meaning. The Bouck decisions are not a universal regulatory rule for Canadian wealth managers, insurers or lenders. They are a concrete reason to examine what a tool contributes to a claim and who authorizes publication.

Where Compliance Checks Belong

Source constraints, draft checks and human approval can work together. A constraint during generation may reduce the chance of an unsupported claim; it cannot guarantee that no misleading sentence will emerge. The final advertisement, including any automatically generated variants, still needs the review required for that communication.

This is the product design question behind Leadlord: how to connect drafting, supporting material and review decisions within a campaign workflow. Keeping those records together can help a reviewer trace an assertion. It is a workflow objective, not a court-endorsed safe harbour or a promise that using a particular platform satisfies the law.

What to Watch Next

The procedural sequence matters: a March decision on pleading sufficiency was followed by a June dismissal of the state-law class action. Any later appeal, amended pleading or federal securities claim must be assessed on its own record.

The useful question for an advertiser is narrower than whether “AI owns the words.” Can the firm explain the claim’s source, what the tool changed and who approved the version that reached the public? That is a workable review discipline without overstating what these courts decided.