The Character.AI Lawsuit — Is a Chatbot's Output Speech or a Product?

Garcia v. Character Technologies is a lawsuit filed on 22 October 2024 in the US District Court for the Middle District of Florida by Megan Garcia, whose 14-year-old son, a user of the AI chatbot service Character.AI, died by suicide in February 2024. The defendants were the operator, Character Technologies; its founders, Noam Shazeer and Daniel De Freitas; and Google. In May 2025 the court declined, at that stage, to accept the defendants' argument that the chatbot's output was speech protected by the First Amendment, and let most of the claims proceed. It was an early, closely watched case asking directly whether an AI's output is "speech" or a "product". The case ended on 7 January 2026 after the parties notified the court of a resolution.

What follows is based on the plaintiff's allegations as summarised in the court's order and on public statements by the company and regulators. Most of the allegations were never tested at trial.

The claims

According to the court's May 2025 order, the plaintiff alleged the following. Character.AI is an app for chatting with "Characters" that portray real or fictional people, and before August 2024 it was rated as suitable for ages 12 and up. Sewell Setzer III began using it in April 2023, mainly talking to Characters imitating figures from a popular television drama, including romantic and sexual exchanges. Within months, the complaint said, he became withdrawn and his schoolwork declined; he saw a therapist who did not know he was using Character.AI.

The claims included product liability (design defect and failure to warn), negligence, wrongful death, violation of Florida's Deceptive and Unfair Trade Practices Act and unjust enrichment. On Google, the complaint pointed to the founders' earlier work on large language models at Google and to a 2 August 2024 deal in which Character Technologies granted Google a "non-exclusive license" for $2.7 billion and Google rehired the founders, and argued that Google shared responsibility.

The company's response

On 22 October 2024, the day the suit was filed, Character.AI published safety updates: a pop-up directing users to the National Suicide Prevention Lifeline when they entered certain phrases about self-harm or suicide, model changes for users under 18, a revised disclaimer on every chat reminding users that the AI is not a real person, and a notification after an hour-long session.

On 12 December it announced a separate model for teenagers, giving them a different experience from adults. It added parental insight into time spent and most-used Characters, and extra language on Characters calling themselves "psychologist", "therapist" or "doctor" telling users not to rely on them for professional advice.

The May 2025 ruling: is the output speech?

The defendants moved to dismiss, arguing that Character.AI's output is speech protected by the First Amendment and that users have a right to receive it. In an order dated 21 May 2025, Judge Anne C. Conway accepted that Character Technologies could assert its users' First Amendment rights. But on whether the output is speech, she drew on Justice Barrett's concurrence in a Supreme Court case suggesting that AI relying on large language models might not reflect a human's expressive choice, and wrote:

the Court is not prepared to hold that Character A.I.'s output is speech.

This was not a final judgment, only a decision not to throw the case out. Still, the fact that at least this court did not, at the outset, let an AI company escape liability by calling its output "speech" drew attention as a point that could shape similar suits.

Regulators and states

In its 29 October announcement, Character.AI said it would remove the ability for users under 18 to engage in open-ended chat with Characters, cutting their daily limit from two hours downward during the transition. It said it would roll out age assurance combining an in-house model with third-party tools including Persona, and would establish and fund an independent non-profit, the AI Safety Lab, focused on safety for AI entertainment. The company said the decision followed feedback from regulators, safety experts and parents.

On California's SB 243, Decrypt reported that earlier provisions such as third-party audits were dropped amid industry pressure, and that Common Sense Media and other groups withdrew their support.

Bankruptcy and users' data

The litigation weighed on the company itself. On 12 October 2026 Character Technologies filed a voluntary Chapter 11 petition in Delaware (Case No. 26-11603). According to KQED, chief executive Colin Adams cited in a sworn declaration the company's high legal costs, its struggle to become profitable and potential investors' concerns over litigation, and told the court that 46 states and territories were investigating it. Disney has offered $20 million to license some of the company's technology, a deal that leaves out users' data and chats.

KQED reported that the service has roughly 12 million monthly users, including at least 2 million children. Citing the trade publication PPC Land, it said attorneys general from 39 states and the District of Columbia filed conditions they want any buyer to meet, including real age verification, removal of chatbots posing as therapists, doctors or lawyers, and limits on using chats to train AI. California Attorney General Rob Bonta and 22 other states separately asked the court to write privacy and safety protections for children into any sale. Conversations that children had confided to chatbots had become assets that could be sold in bankruptcy.

Social impact

The case carried one question—how to protect minors on services where AI plays a friend or partner—into three arenas at once: the courts, regulators and the company's own policies. The result was that the company ended open-ended chat for teens, the FTC opened an industry-wide inquiry, and California passed a law.

There were costs as well. By the company's own account, most of its teenage users had used it creatively within its rules, and they lost its main feature. The family, for its part, brought to court the question of whether product liability can apply to an AI's output, but the case ended in a resolution rather than a verdict, leaving that question to other lawsuits.

The wider story is on the AI timeline; see also large language model. For the earliest chatbot and the first person to write about people confiding in one, see Joseph Weizenbaum.