Suing an AI Company: How Chatbot and Social Media Harm Cases Actually Work

Until recently, the idea of suing a technology company because its product harmed a child seemed close to hopeless. Section 230 stood in the way, causation was hard to prove, and the defendants had limitless resources. That has changed — not because the obstacles disappeared, but because plaintiffs found a different way in.

The Cases That Opened the Door

Two matters are widely regarded as the turning point.

Garcia v. Character Technologies, filed in the Middle District of Florida, was brought by the mother of Sewell Setzer III, a 14-year-old who died by suicide in 2024 after months of interaction with a Character.AI companion. The complaint alleges the persona sent him emotionally and sexually abusive messages. Character.AI moved to dismiss on First Amendment and Section 230 grounds. The court declined to dismiss the case in full at that stage, and the ruling drew wide attention as an early indication that courts may not extend to AI-generated output the same broad protection they have given to content posted by users. The case remains in litigation and the company has contested the allegations.

Raine v. OpenAI, Inc., filed in San Francisco County Superior Court in August 2025 by Edelson PC and the Tech Justice Law Project, was the first wrongful death case brought directly against OpenAI. It is the case that gave Adam's Law its name. OpenAI has disputed the allegations, contending among other things that the teen circumvented its guardrails and was directed to crisis resources many times, and has said its safeguards can become less reliable during lengthy conversations.

Both matters remain contested. Nothing here should be read as a conclusion about either one. But they established that these claims can survive the threshold motions that used to end them, and a substantial volume of similar litigation has followed against OpenAI, Character Technologies, and Google. California state court cases against OpenAI have been coordinated before a single judge, and there have been reports of settlements in some of the Character.AI matters.

The Legal Theories

These cases are deliberately modeled on the frameworks that worked against tobacco and, more recently, against social media platforms. The main theories:

Product liability — design defect

The claim is that the product itself was defectively designed: it was built to maximize engagement and emotional attachment without adequate safeguards for vulnerable users. Framing an AI system as a product rather than a publisher is the central strategic move, because product liability law asks whether the thing was built safely, not whether someone said something harmful.

Failure to warn

A related theory: the company knew or should have known of specific risks — that safeguards degrade over long conversations, that the product tends toward sycophancy, that minors were using it without meaningful age verification — and failed to warn users or parents.

Negligence

The ordinary duty of reasonable care. In California, AB 2 now attaches heightened statutory damages to negligence claims against social media platforms that injure children, which materially changes the economics of bringing one.

Statutory claims

Adam's Law's private right of redress, where the harm involves crisis referral failures, parental notice failures, stripped default protections, or the prohibited categories of harmful output. Consumer protection statutes like California's Unfair Competition Law may also apply.

The Defenses

Anyone considering one of these cases should understand what they are up against.

Section 230. Defendants argue they are interactive computer services immune from liability for content. Plaintiffs respond that AI-generated output is not third-party content — the company's own system produced it — and that design claims aren't about content at all. This is unsettled.

The First Amendment. Defendants argue chatbot output is protected speech. Courts have not uniformly accepted this, and the question of whether machine-generated text carries the same protection as human expression is genuinely open.

Causation. The hardest issue in every case. Defendants argue that mental health crises have many causes and that no one can attribute an outcome to a chatbot conversation. Plaintiffs counter with the products' own logs, internal risk assessments, and safety telemetry.

User conduct. Defendants often argue the user circumvented safeguards. How much weight that carries where the user was a minor, and where the product was designed to be engaging, is contested.

What These Cases Turn On

Evidence, overwhelmingly. The strongest cases have been built on:

• The conversation logs themselves — what the product actually said, and when
• What the company's own systems detected. Allegations in this space have focused heavily on whether monitoring systems flagged concerning conversations and what, if anything, the company did about it.
• Internal documents about safety testing, known risks, and launch decisions
• Product design records showing engagement optimization choices
• Contemporaneous evidence of harm — medical records, school records, family observations

If Your Family Has Been Affected

The single most important thing is preservation. Digital evidence is fragile:

• Do not delete the account. Deleting it may destroy the conversation history permanently.
• Screenshot and export everything you can access, immediately. Many platforms offer a data export; request it.
• Preserve the device. Don't wipe or reset it.
• Request your data formally. Under the CCPA, California residents have a right to know what personal information a business holds about them. That right can reach material the company might not otherwise produce.
• Write down the timeline while it is fresh — when use began, what changed, what you observed.
• Get advice early. These cases involve short preservation windows, complex coordination with existing proceedings, and defendants who litigate hard.

The Bottom Line

This is among the fastest-moving areas of American law. Rulings that seemed unlikely two years ago have arrived; a comprehensive California statute now exists where there was none; and the theories being tested borrow from playbooks that eventually succeeded against other industries. The law is unsettled and the defenses are serious — but the door that was closed is open, and families who preserve their evidence are in a far better position than those who wait.