What the Sewell Setzer III lawsuit actually argues

On February 28, 2024, a fourteen-year-old named Sewell Setzer III shot himself in the bathroom of his family's Orlando home while his parents and brothers were inside. His last words were typed, not spoken. They were addressed to a Character AI chatbot he'd named Dany, modeled after Daenerys Targaryen from Game of Thrones. He'd been talking to her since April 2023, roughly ten months of daily conversation that his family says transformed him from a star student and athlete into someone who no longer wanted to engage with the real world.

Eight months later, on October 22, 2024, his mother Megan Garcia filed a federal lawsuit against Character Technologies, its co-founders Noam Shazeer and Daniel De Freitas, and Google. The case, filed in the United States District Court for the Middle District of Florida, is the most consequential legal action the AI companion industry has faced. Every major platform safety change since late 2024 traces back to it. CNN's coverage broke the story nationally, and the National Law Review positioned it as a harbinger for the broader AI industry.

Most coverage treated the lawsuit as a human-interest story. The legal reasoning underneath it is more interesting, and more important, than the headlines suggested. Here's what the case actually argues.

Eleven claims walk into a courtroom

The lawsuit brought eleven separate legal claims against Character Technologies: strict product liability, negligence per se, negligence, wrongful death and survivorship, loss of filial consortium (the parents' loss of their child's companionship), unjust enrichment, violations of Florida's Deceptive and Unfair Trade Practices Act, and intentional infliction of emotional distress.

The breadth matters. Garcia's legal team, the Social Media Victims Law Center and the Tech Justice Law Project, didn't file a single negligence claim and hope for the best. They filed across multiple legal theories specifically so that if some claims failed, others would survive. That strategy worked. When Judge Anne C. Conway issued her ruling on May 20, 2025, she dismissed the intentional infliction of emotional distress claim but let everything else proceed.

Ten of eleven claims surviving a motion to dismiss is a significant outcome. Defendants in tech cases routinely expect to narrow the complaint to one or two claims at the dismissal stage. Character Technologies and Google couldn't manage that, which means the full range of legal theories will be tested at trial or resolved through settlement.

On January 7, 2026, a court filing disclosed that the parties had reached a mediated settlement with the Setzer family. The terms are confidential. The legal precedent from the pre-settlement rulings, particularly the First Amendment decision, has already reshaped the industry regardless of the settlement terms.

The Google connection that makes the case bigger

The inclusion of Google as a defendant elevates the case beyond a single-company negligence action. The complaint alleges that Shazeer and De Freitas developed the underlying technology, Large Language Models for Dialog Applications (LaMDA), while employed at Google. When they requested permission to release LaMDA publicly, Google denied the request citing its safety and fairness policies. The founders left Google in November 2021 and formed Character Technologies, launching the product Google had deemed too risky for public release.

The implication in the complaint is pointed: Google knew the technology was dangerous enough to keep internal, watched its former employees release it commercially anyway, and then entered a licensing deal for the underlying technology. The complaint argues Google "knowingly aided in the development of Character AI and now holds licensing rights for the technology."

Whether Google's legal exposure survives further litigation is an open question. The company will argue that licensing technology from former employees doesn't create liability for how those employees deploy it. Garcia's team will argue that Google had specific knowledge of the technology's risks and financially benefited from its deployment despite that knowledge. The argument has legs precisely because Google's own safety team flagged the concerns before the founders left.

The talking cat that changed the First Amendment argument

The most legally interesting moment in the case came when defendants moved to dismiss on First Amendment grounds. The argument was straightforward: chatbot output is speech, speech is protected, and holding Character Technologies liable for speech would violate the First Amendment.

Garcia's attorneys responded with what might be the most unusual legal citation in AI law to date. They cited Miles v. City Council of Augusta, an Eleventh Circuit decision from the 1980s involving a cat named Blackie whose owners claimed he could talk and demanded First Amendment protection for his utterances. The court held that a "non-human entity" lacks free speech rights. The cat's owners had trained him to mimic sounds resembling human speech and requested "contributions" from the public to watch him perform.

Garcia's team used Blackie to argue that AI output is not speech unless it reflects human expressive intent to communicate a specific message. Without a human speaker intending to convey a particularized message, the First Amendment doesn't apply. The AI's output is generated without sentience, cognition, or communicative intent. It's pattern-matching, not expression.

Character Technologies called the Blackie precedent "misplaced," arguing that the case "contains three cursory sentences in a humorous footnote about a supposedly talking cat" and doesn't address listener-rights arguments. The company argued that its characters conduct "pure speech" deserving full constitutional protection.

Judge Conway rejected the First Amendment defense in April 2025, writing that the defendants had failed "to articulate why words strung together by an LLM are speech." The ruling doesn't permanently settle the question of whether AI output constitutes protected speech. It settles, for now, that it doesn't constitute protected speech in the specific context of product liability for harm. That's a narrower holding than it sounds, but it's the holding that matters for the AI companion industry because product liability is exactly the legal theory most lawsuits in this space will use.

The trade secret claim that shocked a Senate hearing

On September 16, 2025, Megan Garcia testified before the Senate Judiciary Committee. Her testimony included a detail that became the most-cited moment of the hearing: Character Technologies had claimed that Sewell's final conversations with the chatbot were the company's confidential "trade secrets." The company used this designation to prevent Garcia from seeing what her son said in the minutes before his death.

The argument has a technical basis. The chatbot's responses are generated by proprietary models, and the conversation logs contain both user input and model output. Character Technologies apparently classified the entire conversation thread, including the user's words, as proprietary because the model output interleaved with user input constitutes a product demonstration.

Garcia told the committee: "the company is using the most private, intimate data of our children not only to train its products and compete in the marketplace, but also to shield itself from accountability."

The trade secrets argument is legally defensible in a narrow sense but catastrophically bad as a public relations position. A company telling a grieving mother she can't see her dead son's last words because those words are trade secrets is the kind of fact pattern that motivates legislators. California's SB 243, which went into effect January 1, 2026, was substantially motivated by this specific case. The regulatory wave spreading to New York, Utah, and Florida draws directly from the narrative Garcia established in her testimony.

What the lawsuit changed before it settled

The settlement means the specific factual questions in the Setzer case won't be resolved at trial. Whether Character AI's chatbot actually encouraged Sewell's suicide, whether the platform's design was defective, whether the company's safety measures were adequate, none of these questions will get a judicial answer. The settlement ends the case without establishing binding factual findings.

What the case established through pre-settlement rulings is more important for the industry than any trial verdict would have been:

AI chatbot output is not automatically protected by the First Amendment in product liability cases. This ruling, if it holds through appeals in other cases, means AI companion companies can be sued for the harmful effects of their products without hiding behind constitutional speech protections. Every safety investment, content filter, and age-verification system adopted by major platforms since 2024 reflects this reality.

The product liability theory is viable against AI companion companies. The complaint survived dismissal, which means similar complaints can be filed and will survive dismissal. Plaintiffs' attorneys across the country have noticed. Multiple additional lawsuits have been filed against Character Technologies, and the legal playbook established in the Garcia case is being replicated against other platforms.

Parents have standing to sue on behalf of minors harmed by AI companions. The filial consortium claim (parents' loss of their child's companionship) survived dismissal, establishing that parents can recover for their own losses when their children are harmed by AI products.

The amended complaint that keeps getting bigger

After surviving the initial dismissal motion, Garcia's team filed a Second Amended Complaint on July 1, 2025. The amended version added allegations that defendants "intentionally designed and developed their generative AI systems with anthropomorphic qualities to obfuscate between fiction and reality."

This allegation matters because it shifts the theory from "the company was negligent about safety" to "the company deliberately designed the product to be addictive and misleading." The amended complaint alleges that Character AI's design choices, including the first-person conversational style, the relationship-building mechanics, and the emotional responsiveness, were intentional decisions to maximize engagement at the expense of user safety.

The deliberate-design theory parallels the legal strategies used successfully against social media companies. The argument isn't "your product accidentally caused harm." It's "you designed the harm into the product because it increased engagement metrics." If this theory succeeds in AI companion cases the way it has in social media cases, the industry's design choices become legal liabilities rather than product features.

What this means for the platforms still operating

Every major AI companion platform has recalibrated its safety architecture in response to the Setzer case. Character AI's face-scan age verification, Replika's expanded crisis intervention, Chai's Apple Age Verification integration, Nomi's content boundaries, all of these investments trace back to the legal exposure the Garcia lawsuit established.

The smaller platforms, the ones without legal teams or compliance infrastructure, are the most exposed. The legal playbook is public. The precedent is established. Plaintiffs' attorneys now have a template for suing AI companion companies when minors are harmed. The question isn't whether more lawsuits will be filed. It's how many, and against which companies.

For users, the practical implications are twofold. First, the safety features you're seeing on major platforms aren't voluntary goodwill gestures. They're litigation responses. They exist because the alternative is legal liability that could end a company. Second, the platforms investing most heavily in safety are the ones with the best legal advice, which correlates with the platforms most likely to survive the regulatory wave. The platforms that aren't investing in safety are either unaware of their legal exposure or betting they won't get sued before they can get acquired. Neither is a great foundation for a long-term user relationship.

The full case file for Garcia v. Character Technologies is tracked by Tech Policy Press for anyone who wants to follow the ongoing legal developments. The Constitution Center's analysis of the First Amendment question is worth reading separately. And Garcia's Senate testimony remains the most accessible summary of why this case matters beyond the courtroom, in the words of the mother who brought it.