AI's Next Legal Battleground: Product Liability

British Columbia sued OpenAI last week, alleging the company failed to warn police before February's mass shooting at Tumbler Ridge Secondary School — an attack that killed eight people and wounded 27 others.

The case centers on the loss of life at the school. But at its core lies an argument that plaintiffs are making with growing frequency in lawsuits against AI companies: the tools these firms produce aren't just services that somebody happened to use. They're products — and potentially defective ones at that.

The province alleges OpenAI failed to alert law enforcement after the shooter's conversations with ChatGPT triggered internal concerns. OpenAI has said it maintains measures designed to prevent harm and cooperates with law enforcement when it identifies credible threats.

British Columbia is far from alone in taking AI firms to court. Matthew and Maria Raine sued OpenAI last year after their 16-year-old son Adam died by suicide. They allege ChatGPT validated suicidal thoughts and supplied information about methods of self-harm. OpenAI disputes responsibility for the incident, though it has acknowledged that its safeguards can become less reliable over long conversations.

The Raines' argument mirrors the one made by Megan Garcia, who sued Character.AI and Google after her 14-year-old son Sewell Setzer III died by suicide in 2024. A federal judge allowed significant parts of Garcia's case to proceed, including product liability claims aimed at the design of Character.AI rather than simply the content it generated. Character.AI and Google settled Garcia's case and some related lawsuits in January. The terms were not disclosed.

From service to product

Why are so many of these cases reaching courtrooms? One reason is simply scale. "It's not just a small segment of early adopters who are using it, but now it's reached into the mainstream," says Eric Goldman, associate dean for research and professor of law at Santa Clara University School of Law. Wider use creates more opportunities for things to go wrong — especially among later adopters who may be less informed about the technology's strengths and limitations.

But that doesn't fully explain the volume of cases moving through the courts. Plaintiffs' lawyers are increasingly framing chatbots as products whose features can be negligently designed. In the Garcia case, Judge Anne Conway said Character.AI could be treated as a product because Garcia's claims concerned defects in the app's design rather than its output.

The ruling did not find that Character.AI caused Setzer's death — something the firm is at pains to point out. It does, however, show a legal system willing to debate these cases on the basis of defective products rather than other theories of liability.

Eran Kahana, an AI and cybersecurity lawyer and fellow at Stanford Law School, expects the focus on design to intensify. "What we have here is fundamentally and irretrievably a chatbot design problem," he says. Features such as memory, emotional mirroring, and sycophancy are core to the current crop of chatbots because they make the systems feel more human. But they can also put vulnerable users at risk, Kahana argues.

Beyond lives lost

These arguments aren't limited to cases involving death. Nippon Life Insurance Company of America sued OpenAI in Illinois in March, alleging ChatGPT effectively practised law without a licence after a woman used it while trying to reopen a settled disability case. The complaint says ChatGPT helped her with legal arguments and court filings. OpenAI has moved to dismiss the case. Kahana sees it, too, as an example of a system whose design can allow it to cross a line its developer arguably should have anticipated.

It's tempting to blame the technology alone. Goldman notes these issues sit within a broader set of social problems. "So many of the concerns about generative AI relate to things like mental health issues, and mental health issues are not a generative AI problem," he says.

Unsettled questions

Courts still haven't answered the questions that could dictate the success or failure of many more cases: whether a chatbot is a product, and how much responsibility its maker bears if people act on its advice. Kahana argues existing product liability law may provide the answers. Goldman sees another path in past precedent: lawmakers could eventually grant generative AI companies statutory protection from liability, echoing the shield internet companies received in the 1990s under Section 230.

That would prove controversial, but it might be preferable to precedents set case by case. Goldman cautions against expecting litigation to fix everything the technology exposes. "We can fix generative AI," he says, "and might not fix any problem."

For AI developers, the stakes are now structural. If courts keep accepting the product-liability framing, design choices — memory, emotional mirroring, sycophancy — become legal exposure, not just engineering tradeoffs.