In February 2026, a California court did something procedural and, on its own, not newsworthy: it coordinated a group of pending lawsuits so one judge could handle shared pretrial questions instead of a dozen separate courtrooms doing it separately. That's normal case management. What makes it worth writing about is what the cases have in common.
TL;DR: The Superior Court of California, San Francisco County, coordinated at least a dozen wrongful death and product liability cases against OpenAI into a single proceeding, In re: ChatGPT Product Liability Cases, JCCP No. 5431, in an order entered February 3, 2026. More cases have since been added, including one naming CEO Sam Altman personally. Coordination is procedural, not a finding of liability, and trial is years away. The real governance question it raises now: does your chatbot have a deliberate, tested response when a user discloses a crisis, or are you relying on default model behavior and hoping.
What was actually coordinated
The proceeding is formally titled In re: ChatGPT Product Liability Cases, JCCP No. 5431. The cases share a common thread: each alleges that ChatGPT's design contributed to serious harm, including wrongful death, by reinforcing a user's distress rather than interrupting it. Legal theories across the coordinated cases include strict product liability and negligence for defective design, failure to warn, and wrongful death.
Among the plaintiffs are Matthew and Maria Raine, whose case (Raine v. OpenAI) was among the first and best documented. More families have filed since the original coordination order, including a case brought by a Canadian mother whose 24-year-old daughter died by suicide in July 2025, after having disclosed suicidal thoughts to ChatGPT repeatedly over roughly a year and a half. That suit names Sam Altman personally as a defendant, alleging he pushed a product to market on a compressed testing timeline.
It is worth being precise about what coordination means and doesn't mean. A JCCP order is a case management tool. It does not decide whether OpenAI is liable for anything, and none of the underlying allegations have been tested at trial. These cases are in early stages, and resolution, if any of them reach a verdict at all rather than settling, is likely years out.
Why this matters even if you don't build chatbots
This site has covered Rhode Island's new law requiring AI companion products to detect suicidal ideation and route users to crisis resources, backed by penalties up to $15,000 a day. Read next to this litigation, that law stops looking like a hypothetical precaution and starts looking like a direct response to exactly this fact pattern: a chatbot that was in a position to redirect someone in crisis, and, according to the allegations, didn't.
That's the actual governance lesson, independent of how any individual case resolves. If your product, or a product you've built on top of a general-purpose model, is in a position where a user might disclose a mental health crisis, self-harm intent, or a request for information that could facilitate harm, "we used the model's default behavior" is not the same thing as "we made a decision about how our product handles this." Courts and legislators are both, right now, treating that distinction as the one that matters.
What to actually do with this
If you're deploying a general-purpose model in any customer-facing context, know what the underlying model does by default when a user discloses a crisis, and don't assume it's adequate for your specific use case. A customer support bot for a SaaS product has a different risk profile than a companion or wellness-adjacent product, but "we didn't think about it" is the position every defendant in this litigation is currently trying to avoid being in.
Document the decision, not just the outcome. If your team evaluated crisis-response behavior and decided your product's exposure is low, write that down, with the reasoning, when you make the call. The cases in JCCP 5431 turn substantially on whether design choices were deliberate and whether risks were foreseeable. A dated record showing you considered the question is a materially different position than reconstructing your reasoning after something goes wrong.
Revisit this if your product's user base or use case changes. A tool built for internal, professional use carries different exposure than the same tool opened up to a general consumer audience. The moment your users are more likely to include people in genuine distress, the standard of care implicitly being litigated here starts applying to you too.
Related Reading
- Rhode Island Just Passed 3 AI Laws With a $15,000-a-Day Penalty
- AI Liability Insurance: What Coverage Actually Exists in 2026
- Agentic AI Liability: Who Is Actually Responsible
- AI Vendor Due Diligence Checklist for Small Teams
- AI Incident Response Plan: What Regulators Expect
Sources: Hunton Andrews Kurth, "California Superior Court Consolidates Product Liability Actions Against OpenAI", National Law Review, CBS News, Social Media Victims Law Center.
