The People v. AI
When a chatbot is the proximate cause of harm, who is responsible?
For two years, the AI liability conversation lived almost entirely in law review symposia and white papers. The argument was theoretical: existing tort law, product liability doctrine, agency law, and statutory frameworks were all built for a world where harm flowed from human action, and they did not yet know what to do with a system that could act on its own.
That conversation has moved out of the symposium and into court filings. In the past nine months, families have begun suing AI companies for wrongful death, addiction, psychosis, and mass-casualty harm. Plaintiffs' firms have organized. State attorneys general have filed. Congress has held hearings. The institutional pretense that AI is "just software" - and therefore not the subject of product liability - is being tested for the first time, in cases that will define the next decade of the industry.
• The wrongful-death docket against OpenAI is now substantial and growing. Lawsuit Informer reports at least four major U.S. civil actions filed against OpenAI as of May 2026, covering wrongful death, self-harm, and mass-casualty harm. The OpenAI litigation landscape moved from theoretical to active in the second half of 2025 and accelerated in the first half of 2026. Most filed cases plead strict products liability, negligence, and wrongful death theories rather than treating chatbot output as user-generated content. California has emerged as the dominant venue, with state and federal cases concentrated in San Francisco. (lawsuitinformer.com)
• The foundational case in the new docket was filed in August 2025. Matthew and Maria Raine filed suit in San Francisco County Superior Court against OpenAI, CEO Sam Altman, and unnamed employees and investors, alleging that ChatGPT contributed to the suicide of their 16-year-old son Adam Raine. The complaint pleaded seven causes of action anchored on California strict products liability. In October 2025, the Raines amended their complaint to allege intentional misconduct, citing internal OpenAI policy documents (the Model Spec) that they allege show the company made conscious decisions to remove longstanding safety protocols. The amendment opened a pathway to punitive damages.
• The mass-casualty cases arrived in April and May 2026. Seven federal lawsuits were filed against OpenAI and Sam Altman in California federal court by families affected by the February 2026 Tumbler Ridge school shooting. The complaints allege negligence, wrongful death, and product liability, including a specific allegation that the account-enforcement system was circumventable. Plaintiffs' counsel Jay Edelson has signaled the lawsuits are part of a broader effort. On May 10, the family of Tiru Chabba, killed in the April 17, 2025 mass shooting at Florida State University, filed an eight-count federal complaint that included a novel negligent entrustment count.
• The medical-advice frontier opened on May 12. Texas residents Leila Turner-Scott and Angus Scott filed suit in California state court alleging that ChatGPT advised their 19-year-old son Sam Nelson that it was safe to combine kratom and Xanax. The combination of those drugs and alcohol resulted in Nelson's death in May 2025. The complaint brought defective design, failure-to-warn, negligence, and wrongful death claims, plus a claim under California's Unfair Competition Law and one under a state law provision prohibiting AI from representing itself as a licensed health practitioner. The complaint alleges that ChatGPT "engaged in the unlicensed practice of medicine" by making personalized dosing recommendations. The lawsuit also asks the court to pause OpenAI's rollout of ChatGPT Health. (news.bloomberglaw.com)
• California has already foreclosed the most anticipated defense. Reuters reporting on the Turner-Scott complaint highlighted a passage from the filing that quotes a 2026 California law directly: "In California, if plaintiffs prove they were harmed by defendants' AI-powered product, defendants will be liable for that harm, no matter how clever, independent, willful, spiteful, uncontrolled, rebellious, free-spirited, libertine, stochastic, or autonomous the beast they have birthed may be." The statutory text matters more than the rhetorical flourish: California enacted legislation that took effect January 1, 2026, providing that a defendant facing liability for harm caused by an AI system cannot use the system's autonomous operation as a defense.
• By the end of 2025, the docket already included at least ten cases. A May 2026 Psychiatric Times preview reported that by the end of 2025, there were at least 10 known lawsuits against OpenAI and Character Technologies with allegations including wrongful death, involuntary manslaughter, sexual abuse, negligence, and product liability. These cases involve six adults and four minors, seven of whom died by suicide. Plaintiffs argue that chatbot makers are responsible for encouraging suicides - even acting as a "suicide coach" - and driving some users to psychosis or to commit murder. The OpenAI cases allege that the outcomes were both "predictable and the result of deliberate design choices." (psychiatrictimes.com)
• State attorneys general have joined the docket. In early 2026, Character Technologies was sued by the state of Kentucky for "harmful, explicit, and psychologically dangerous" content directed at minors. The Kentucky AG called it the first state action against an AI chatbot company "that has preyed on children and led them into self-harm." Parents of alleged victims testified at a Senate Judiciary Hearing in September 2025 titled "Examining the Harm of AI Chatbots." Since then, legislation has been introduced or passed restricting chatbot use by minors, requiring that chatbots clarify they are not human, and prohibiting chatbots from acting as psychotherapists. Illinois and Utah banned therapeutic bots.
• The product-versus-content framing question is the threshold fight in every case. Lawsuit Informer's analysis identifies it cleanly: the product-versus-content framing question is the first real fight and will decide whether AI companies can be sued for chatbot output at all. If chatbot output is "content" - speech produced by users through a tool - then Section 230 and First Amendment defenses apply, and AI companies are largely immunized. If chatbot output is a "product" - output generated by a machine designed and sold by the company — then strict product liability applies, and the analysis runs through design defect, manufacturing defect, and failure to warn. The Garcia v. Character Technologies decision in 2025 signaled that at least one federal court is willing to entertain product liability claims against AI chatbot companies.
• The agentic AI liability question is the next frontier - and the gap is wider. A Clifford Chance analysis frames the structural problem precisely: the laws of agency and vicarious liability require there first to be a human agent or employee who is primarily responsible for the harm, before their employer can be held responsible. With a truly autonomous AI agent, there may be no human "employee" acting at the moment of harm - the AI acts on its own algorithmic "will." Courts and commentators have consistently noted that without a human "agent," vicarious liability fails by definition. Scholars now refer to this as the "AI responsibility gap." (cliffordchance.com)
• The University of Chicago Law Review proposed the cleanest response to the gap. A recent article argued that the best solution is to employ objective standards familiar in many parts of the law. The AI programs themselves are not the responsible actors; they are technologies used by humans that have effects on other humans. The real question is who should be held responsible for the use of AI and under what conditions. Because AI agents lack intentions, the law should hold them - and the people and companies that employ them - to objective standards: negligence, strict liability, or the highest level of care when the AI performs the functions of a fiduciary. The principle: people should not be able to obtain a reduced duty of care by substituting an AI agent for a human agent.
• The deploying enterprise is becoming the locus of liability. A MindStudio analysis from April crystallized the current consensus: AI liability is not assigned to the AI - it sits across a stack of human actors: model developers, platform vendors, deploying organizations, and end users, in varying proportions depending on where the failure occurred. The deploying organization carries the heaviest load. Enterprises that grant AI agents authority to act on their behalf are generally treated as principals responsible for their agents' conduct. The Mobley v. Workday matter - in which an applicant alleged Workday's algorithmic screening discriminated against him across more than 100 applications - was the early signal that AI vendors could face direct liability for actions taken on behalf of their customers.
• The insurance industry is now pricing the gap. Coverage products for AI liability are emerging, but underwriting remains immature. Most current AI-deployment contracts allocate liability through indemnification provisions, scope-of-authority clauses, and disclaimers - but those allocations cannot override statutory obligations or eliminate basic tort duties owed to third parties. The compliance burden falls heaviest on enterprises that have deployed agents without robust audit logs, approval thresholds, or named human accountability.
Orthogonal Take
The AI liability story is not one story. It is at least three, running on different timelines, with different defendants, different harms, and different doctrinal frameworks.
The first is the consumer chatbot story. Wrongful death, suicide, addiction, psychosis, medical-advice harm, and mass-casualty cases brought by individuals against OpenAI and Character.AI. The cases test whether chatbot output is "content" (immunized by Section 230 and the First Amendment) or "product" (subject to strict liability). California has gone the furthest in foreclosing the most attractive industry defense, by statutorily eliminating the "the AI did it on its own" argument as a shield. The other forty-nine states have not. The federal courts are still working it out.
The second is the autonomous-agent story. When an AI system books a flight, sends a message, executes a trade, files a document, or signs a contract, who is bound? Who is liable when it goes wrong? The doctrinal frameworks built for human agents - agency, vicarious liability, respondeat superior - were all built around the assumption that there is a human in the loop. That assumption no longer holds for the systems being deployed in 2026.
The third is the deployer story. Enterprises that deploy AI agents are increasingly being treated as principals - responsible for the conduct of agents they have granted authority. The Mobley matter is the leading edge of what will become a much larger category. The bet for now is that contractual indemnification, careful scope-of-authority design, and audit logs will be enough to manage exposure. Courts have not yet ruled on whether they are.
The deeper point - the one easiest to lose in the case-by-case coverage - is that the AI industry has been operating for two years on a set of assumptions about its own liability that are not going to survive contact with the legal system.
The assumption that chatbot output is content, not product. The assumption that user agreement disclaimers eliminate duty of care to third parties. The assumption that "the model did it" is a defense. The assumption that mental health harm from chatbot interactions is too attenuated for standing. The assumption that autonomous-agent harm has no proximate cause sufficient to support liability.
Each of those assumptions is being tested simultaneously, in different cases, in different jurisdictions, with different plaintiffs, against different defendants. Most of them are going to fail.
What replaces them is the part nobody has yet built.