When an AI Tells You to Die: Is a Chatbot a Product, a Service, or Speech?

Credit: Stock Images

A defective toaster short-circuits and burns you; a rickety park bench splinters and scratches you; a newspaper promotes misinformation. The common denominator: American law knows exactly how to treat each scenario. But what happens when a commercial product, designed to speak, interact, remember, and build a relationship with you, persuades you to harm yourself? An influx of court cases concerning interactions between users and generative AI chatbots that have allegedly reinforced delusional beliefs, fostered emotional dependency, and encouraged impulsive behavior resulting in, in extreme cases, fatal harm, has brought this question to the limelight. These cases force courts to answer a question tort law was never designed to address: when software causes harm through language, is it simply a defective product, or protected speech?

Existing legal categories break down quickly when classifying AI products. Plaintiffs argue that AI systems are commercial products with defective designs or inadequate warnings, while developers argue that users interact with a software service rather than a conventional product. More broadly, because the identifiable mechanism of injury is speech, liability could implicate the First Amendment. In short, all of these viewpoints are correct, and at the same time all are wrong: each characterization captures just part of how AI chatbots function, but none maps neatly onto the technology as a whole.

Whether courts characterize an AI chatbot as a product, service, or form of expression is not just semantic; it's a real legal classification that will shape the legal duties developers owe users, what plaintiffs must prove, and what defenses companies can invoke. In fact, substantial lawsuits against OpenAI are currently being managed through coordinated proceedings in California and Florida, with attorneys general seeking court restrictions on OpenAI’s future model development, arguing that stronger safeguards are necessary to protect younger users. As the litigation landscape expands, the need to establish these labels quickly is growing.

On the product-liability front, courts traditionally ask whether a product creates an “unreasonable” risk and whether a reasonable alternative design could have meaningfully reduced it. That “unreasonable” risk is typically characterized under two types of “defect”: content and design. Under these circumstances, a content defect would hold the issuing company liable for the language it generated, distinguishing between harm arising from specific chatbot responses and harm arising from the broader system design that may have caused those responses. These designs range from engagement mechanisms and anthropomorphic features like voice responses to programmed responses to indications of self-harm. In fact, in 2025, the case Garcia v. Character Technologies involved a federal district court allowing most claims tied to the suicide of 14-year-old Sewell Setzer III to survive dismissal. In its decision, the court held that Character.AI, the AI chatbot, would be treated as a “product” insofar as the claims concerned alleged defects in the application rather than the ideas or expressions it generated. ​

A design defect, by contrast, would allege that the system was built in a way that encouraged harmful interactions, a more product-design-oriented take than the prior. Traditionally, software has posed difficulties for enforcing product liability because liability doctrine was designed around tangible goods. Developers can and have argued that a chatbot is an ongoing, interactive service rather than a fixed physical object. Thus, though more product-centric, this route becomes much harder to invoke, pushing plaintiffs to rely more heavily on other legal doctrines.​

On the First Amendment front, things become even murkier; if the courts impose liability because the chatbot generated certain phrases, companies are provided the foothold of the argument that the government is imposing penalties based on expressive content. But even this argument leaves questions unanswered, including who exactly is speaking? The developer did not manually author the harmful output, the user did not specifically elicit it, and the model itself is not a legal person. In Garcia v. Character Technologies, Character.AI argued that imposing this type of liability would limit users’ right to receive speech. The district court declined to hold that chatbot output constituted protected speech, acknowledging the authorship and ownership question without resolving it. 

This issue has long since expanded beyond families suing chatbot developers. On October 14, 2026, eight states sued OpenAI under state consumer-protection laws. British Columbia has also sued, alleging OpenAI failed to act appropriately after safety systems flagged potentially dangerous chatbot interactions. San Francisco has filed federal lawsuits arising from a school shooting in Canada earlier in 2026. The law is attacking the same underlying problem through several doctrinal routes, ranging from products liability and negligence to consumer protection and constitutional defenses.

Given generative AI sits at such an unusual intersection of tort, software, and constitutional law, courts are just now beginning to enter the fray of determining what exactly these AI systems are, whether they should inherit the liabilities traditionally associated with most other consumer products, and even if they should attain protections associated with freedom of expression. As the litigation landscape expands and diversifies, the question is not which category wins out, but how courts adapt doctrine that was developed before products could influence and hold conversations with the very people using them.

Isaiah Sohn is a junior at Brown University studying Applied Mathematics-Economics. He is a writer for the Brown Undergraduate Law Review and can be contacted at isaiah_sohn@brown.edu.

Nicolas Abril is a senior from Florida studying Applied Mathematics and Economics. He is a Junior Blog Writer with the Brown Undergraduate Law Review and can be contacted at nicolas_abril@brown.edu. 

Alice Kovarik is a junior from St. Louis, Missouri studying International and Public Affairs and Economics. She is an editor for the Brown Undergraduate Law Review blog and can be contacted at alice_kovarik@brown.edu.