
Children's class action asks OpenAI for equity. Arizona's state suit is different
Filed 13 October in federal court in California, the complaint says a minor's "Continue" click is void and seeks a share of a company it values at over $1 trillion. Arizona's attorney general sued a day later in state court on a deception theory.
A new class action asks OpenAI to pay children with a share of the company. It was filed on 13 October in the U.S. District Court for the Northern District of California as Case No. 3:26-cv-11781.
The equity request is the long shot. The main exposure sits in two declaratory counts, which ask a court to say that a child's click on "Continue" never formed a contract. A ruling for the plaintiffs could force OpenAI to rewrite its consent flow without a court reaching the equity claim. The complaint itself values OpenAI at "over $1 trillion", which is what makes the equity claim headline material.
The most unusual line is in the prayer for relief. Plaintiffs ask for "an equity stake in OpenAI, or the cash value thereof, attributable to data that OpenAI wrongfully took from Plaintiffs". The ask is measured in tokens, the units of text ChatGPT processes.
The 81 page document names Jane Doe, who sues for her minor children, MC1 and MC2. The defendants are the OpenAI Foundation, OpenAI OpCo, OpenAI Holdings and OpenAI Group PBC. The children are 12 and 16, live in Hawaii, and began using ChatGPT in February and January 2026.
The proposed class is much wider. It covers everyone in the United States who clicked "Continue" on ChatGPT's Terms of Use while under 18. They must also have given OpenAI at least one token of data from November 2022 to the present. The class is defined by a click, and one token is enough to be a member.
The legal chain starts with the contract. Account creation, the complaint says, needs a name, a claim to be 13 or older and a click on "Continue" (paragraph 3). The second count invokes California Family Code section 6701 to call the Terms of Use void from the start.
The complaint cites Lee v. Hibernia Savings and Loan Society, a 1918 California decision, for the point that a minor who misstates his or her age is still not bound. The first count asks for a declaration under Civil Code section 1568.5 that the parental consent method was ineffective. Without a valid contract, the argument runs, OpenAI had no licence to use what children typed.
The step from there to equity is the novel one. The prayer cites accession under Civil Code sections 1031 to 1033, an old property doctrine about owners of materials that are transformed into something new. It also cites restitution, constructive trust and disgorgement. Applied to model training, the theory treats children's data as an input that became part of the product and asks for a proportionate share.
The complaint adds that "OpenAI already numbers the tokens associated with each person" (paragraph 14). It says the measuring stick therefore already exists inside the defendant.
The pleading is narrower than its prayer. It has three causes of action: the two declaratory counts and unjust enrichment. The prayer reaches to punitive damages and fees but names no damages figure.
The Arizona case follows a different route. Attorney General Mayes filed on 14 October in Maricopa County Superior Court under the Arizona Consumer Fraud Act. Her office says ChatGPT "encourages minors to share personal information exclusively with" it. The class complaint mentions no attorney general.
These are separate cases. One is a private class in federal court, with California statutes and a jury demand. The other is a state suing in its own superior court on a deception theory. They cannot be folded together. A ruling in one binds the other only as persuasive reasoning, if at all.
The class complaint gives no class size, no token count and no valuation beyond "over $1 trillion". Its line that "millions of American children use ChatGPT weekly" (paragraph 95) describes usage, not membership.
The scale is simple arithmetic. At $1 trillion, each 0.01 percent of enterprise value is $100 million. The complaint cites a $5 million floor for federal jurisdiction, and 0.0005 percent would clear it. A claim on a fraction of a basis point of OpenAI is still a large sum, and every token counted raises it.
Settlement leverage would come from three features: the declaratory framing, the reservation of the right to plead 50 state subclasses and a headline ask no court has granted. Leverage still has to survive the pleading as written, and OpenAI's best reply is the obvious one.
Minors can disaffirm contracts. Disaffirmance ordinarily unwinds a bargain and returns what each side received. It does not obviously give a child a property interest in a trained model, and the accession statutes have not been tested against model training.
The cheaper ruling is the declaratory one, and it is the more dangerous for product teams. A court that accepts the section 1568.5 or 6701 theory could force a rewrite of the consent flow without reaching equity. Many consumer AI apps rely on a self reported age and a click. Each would then choose between age assurance at sign up, with its friction and verification vendors, and a parental consent step a court could accept.
Under federal rules, OpenAI's answer or motion to dismiss falls due 21 days after service. If it waives service, the deadline is 60 days after the request is sent. A waived response would then land around 12 December. A motion against the two declaratory counts would be the first test. The ruling on unjust enrichment would then decide whether the equity request is alive past 2027.