The ChatGPT Product Liability JCCP Reconvenes With 23 Cases, No MDL and No Arbitration Fight

The ChatGPT product liability JCCP returns to Department 304 of the San Francisco County Superior Court today, 23 September 2026, for the first case management conference since Judge Ethan P. Schulman installed a plaintiffs' leadership structure on 4 August. In re ChatGPT Product Liability Cases, JCCP No. 5431, now carries roughly twenty-three coordinated wrongful death and personal injury actions against OpenAI — and the three things that define it are all absences: there is no federal MDL, no bellwether track, and no motion to compel arbitration on a docket built entirely out of a consumer contract that contains one.

That last absence is the one worth a practitioner's attention. Every claimant in this proceeding, or the person whose estate they represent, agreed to OpenAI's Terms of Use. Those terms carry a binding individual arbitration clause, a class action waiver and a 30-day opt-out. In almost any other consumer mass tort, the first order of business would be a fight over that clause. Here it has not happened, and the reasons it has not happened are the most useful thing the ChatGPT product liability JCCP has to teach right now.

What is JCCP 5431 and which cases does it cover?

JCCP 5431 is the California state-court coordinated proceeding formally captioned In re ChatGPT Product Liability Cases, Judicial Council case number CJC-25-005431, assigned to Judge Ethan P. Schulman in Department 304 of the San Francisco County Superior Court. It gathers personal injury and wrongful death actions alleging that OpenAI's ChatGPT — and in some complaints Microsoft as a co-defendant — caused psychological harm and death.

The San Francisco Superior Court granted the coordination petition in February 2026 over twelve then-pending actions. The proceeding has since grown by add-on to roughly twenty-three cases and counting, drawn from San Francisco, Los Angeles, Alameda and San Diego counties, with plaintiffs resident across the country. The originating case is Raine v. OpenAI, No. CGC-25-628528, filed in San Francisco on 26 August 2025 by the parents of sixteen-year-old Adam Raine, who allege that GPT-4o deepened their son's suicidal ideation before his death in April 2025. An amended complaint filed in October 2025 added allegations that safety guardrails were deliberately relaxed, opening a path to Punitive Damages.

The liability theory across the coordinated complaints is conventional products law pointed at an unconventional product. Plaintiffs plead that ChatGPT is unreasonably dangerous as designed; that it reinforced delusional beliefs, endorsed suicidal ideation and supplied self-harm information; and that OpenAI released GPT-4o without adequate safety testing despite internal warnings about sycophantic design. The causes of action are Design Defect, Failure to Warn, negligence and wrongful death, with consumer protection counts in some complaints.

What is happening at the 23 September 2026 case management conference?

Today's hearing is a further case management conference in a coordinated proceeding roughly seven weeks old in its organised form: the first conference was held on 24 July 2026, Case Management Order No. 1 followed on 4 August, and 23 September is the next date the court set. It is the first opportunity for the newly appointed leadership to put a pleading and discovery schedule in front of the coordination trial judge, and the same date appears on the Raine docket as that case's next hearing.

What a first working conference in a coordinated proceeding typically has to settle is not glamorous but it is decisive: whether plaintiffs file a master or consolidated complaint against which OpenAI demurs once rather than twenty-three times; a demurrer briefing schedule on the defences everyone expects, including the First Amendment; a protective order and an ESI protocol for a defendant whose most probative evidence is conversation logs and internal model-safety records; a fact sheet regime standing in for the federal Plaintiff Fact Sheet (PFS); and a mechanism for the steady stream of add-on cases. None of that is pre-ordained, and nothing in the public record fixes today's outcome in advance.

How the coordination happened, and why Department 304

California has no Judicial Panel on Multidistrict Litigation (JPML). It has a coordination statute. Under Code of Civil Procedure section 404 and following, a party to one of two or more civil actions sharing a common question of fact pending in different courts may petition the Chair of the Judicial Council to coordinate them. Section 404.1 sets the standard: coordination is appropriate if the actions are complex and, treating the convenience of parties, witnesses and counsel, the relative development of the actions, the efficient use of judicial facilities, the calendar of the courts, the disadvantages of duplicative and inconsistent rulings and the likelihood of settlement, one judge hearing all of them will promote the ends of justice.

The mechanics then run in two steps. The Chair assigns a coordination motion judge to decide the petition; if the petition is granted, the Chair either assigns a coordination trial judge or authorises a presiding judge to make the assignment, under section 404.3. Immediately on assignment, the coordination trial judge may exercise over each coordinated action every power available to a judge of the court in which that action was pending. That is how a San Francisco judge came to control cases filed in Los Angeles, Alameda and San Diego.

The timeline below is the public record of this proceeding to date.

Date	Event
26 August 2025	Raine v. OpenAI, No. CGC-25-628528, filed in San Francisco County Superior Court
October 2025	Amended complaint filed in Raine, adding allegations of relaxed self-harm safeguards
November 2025	Seven further actions filed in California state courts on behalf of four decedents and three survivors
26 November 2025	OpenAI answers in Raine, denying causation and pleading misuse of ChatGPT
December 2025	Lyons v. OpenAI, No. 25-cv-11037-RS, filed in the Northern District of California
February 2026	San Francisco Superior Court grants coordination over twelve actions; JCCP No. 5431 opened
13 April 2026	Chief Judge Richard Seeborg declines to abstain under Colorado River in Lyons
24 July 2026	First case management conference, Department 304
4 August 2026	Case Management Order No. 1 entered, appointing plaintiffs' leadership
23 September 2026	Further case management conference — today

Who is leading the plaintiffs, and what Case Management Order No. 1 actually does

Case Management Order No. 1, entered on 4 August 2026, appointed four plaintiffs' co-lead counsel, one liaison counsel and a six-member plaintiffs' steering committee, one seat of which was left to be designated. The court paired the appointment with a warning that it would restructure leadership if the appointees failed to work together cooperatively — a signal that matters in a proceeding assembled from firms that came in through four different county courthouses.

Three features of the order are worth reading closely.

First, it divides authority rather than centralising it. Co-lead counsel direct liability pleadings and common discovery for the coordinated plaintiffs; each plaintiff's own counsel retains control of that plaintiff's individual action and, critically, of any settlement of it. That is a deliberate choice. It keeps the proceeding from behaving like an inventory and preserves the individual character of claims that will, if they are ever valued, be valued case by case rather than off a Settlement Matrix.

Second, it looks forward to a federal proceeding that does not exist yet. The order directs liaison counsel to coordinate with leadership in any federal multidistrict litigation covering the same subject matter. Courts do not write that sentence idly; it is an acknowledgment that the federal filings are accumulating and that someone will eventually petition the Judicial Panel on Multidistrict Litigation (JPML).

Third, it protects self-represented plaintiffs. The order requires that pro se plaintiffs receive filings, notices and shared discovery access on the same basis as represented parties, and Judge Schulman interlineated by hand onto the signed order that nothing in the leadership section limits the right of any self-represented plaintiff to file papers concerning issues unique to their individual case. Leadership orders routinely squeeze unrepresented litigants out of their own dockets; on a docket growing by add-on, where a grieving family may file before finding counsel, the carve-out has practical teeth.

Is there a federal MDL for ChatGPT lawsuits?

No. As of today there is no Multidistrict Litigation (MDL) docket for ChatGPT personal injury or wrongful death claims, and no JPML docket number has been opened for them. The federal ChatGPT injury cases are proceeding individually before the judges to whom they were assigned, and the only OpenAI MDL in existence is MDL No. 3143, the copyright infringement litigation before Judge Sidney Stein in the Southern District of New York, which has nothing to do with personal injury.

That leaves a two-track structure familiar from talc, opioids and hernia mesh: a California JCCP doing the coordinated work on one side, and unconsolidated federal actions on the other. The federal track is not idle. In Lyons v. OpenAI, No. 25-cv-11037-RS, brought by the administrator of the estate of Stein-Erik Soelberg, OpenAI asked the Northern District of California to stand down in favour of the parallel California state proceeding. On 13 April 2026 Chief Judge Richard Seeborg refused. Applying Colorado River Water Conservation District v. United States, he held that a stay is not appropriate where there is substantial doubt that the state proceedings will resolve the federal case, noting that the state action concerns whether ChatGPT contributed to a killing while the federal action concerns whether it contributed to a suicide, and that negligence and product liability questions about new technology are litigated in federal courts all the time.

For anyone choosing a forum, the difference between the two tracks is worth stating plainly.

	JCCP 5431 (California)	A federal MDL (if one forms)
Authority	Cal. Code Civ. Proc. §§ 404–404.9	28 U.S.C. § 1407
Who decides	Chair of the Judicial Council; coordination motion judge	Judicial Panel on Multidistrict Litigation (JPML), seven judges
Scope	California state actions only	Federal actions nationwide
Judge	Coordination trial judge with all powers of the originating courts	Transferee judge for pretrial purposes
Trial	Coordination trial judge may try the coordinated actions	Remand to the transferor court for trial after pretrial
Adding a case	Add-on request to the coordination trial judge under rule 3.544	Conditional transfer order, subject to objection
Status here	Granted February 2026, active	No petition filed, no docket opened

Can OpenAI force ChatGPT wrongful death claims into arbitration?

Probably not for the wrongful death claims, and that is the most likely explanation for the silence. OpenAI's consumer Terms of Use require users to resolve claims arising out of or relating to the terms by final and binding individual arbitration, prohibit class arbitrations, class actions and representative actions, and allow a user to opt out within 30 days of account creation or of an update. That is a textbook Arbitration Clause paired with a Class Action Waiver, and in ordinary consumer litigation it would be the whole ballgame.

California wrongful death law blunts it. A wrongful death claim belongs to the heirs, not to the decedent; it is their own cause of action for their own loss, not a claim inherited through the estate. A non-signatory heir asserting that independent claim is not bound by the decedent's Arbitration Agreement merely because the agreement purports to bind heirs. Daniels v. Sunrise Senior Living, Inc., 212 Cal.App.4th 674 (2013), makes the point in the elder care context: language binding "heirs, representatives, executors, administrators, successors and assigns" binds those asserting survival claims on the decedent's behalf, but does not reach third parties with claims independent of the survival claims. Survival claims brought by the estate stand differently, because there the estate stands in the decedent's shoes.

So the coordinated proceeding presents a split. Survivor plaintiffs who used ChatGPT and sue for their own injuries are signatories and face the clause directly. Estates asserting survival claims step into the decedent's contractual position. Parents and spouses suing for wrongful death are, on Daniels, asserting claims that are theirs alone. Moving to compel would at best fracture the proceeding rather than end it, and it would hand plaintiffs a public fight about forcing bereaved families into private arbitration — in front of a judge and in front of the attorneys general already circling the company.

There is a second, colder calculation, and it cuts the other way. Waiting is not free. In Uhl v. Roblox Corp., decided 3 September 2026, a divided Ninth Circuit panel held that Roblox had waived arbitration of a parent's child-safety claims by removing the case, litigating a remand motion, moving to dismiss on the merits and waiting roughly eleven months before invoking its terms of service. Since Morgan v. Sundance, Inc., 596 U.S. 411 (2022), a party opposing arbitration no longer has to show prejudice to establish Waiver of Arbitration; knowledge of the right plus conduct inconsistent with it is enough. OpenAI answered in Raine on 26 November 2025 and has litigated the coordination since. Whatever right to arbitrate exists against the signatory plaintiffs is aging in public, and a Motion to Compel Arbitration filed after a demurrer round would be filed into a Ninth Circuit and a California bench that have both grown impatient with the tactic.

What OpenAI has actually pleaded

OpenAI's answer in Raine, filed 26 November 2025, is the fullest public statement of its defence, and it is a comparative fault answer rather than a jurisdictional one. The company pleaded that the plaintiffs' alleged injuries were caused or contributed to, directly and proximately, in whole or in part, by Adam Raine's "misuse, unauthorized use, unintended use, unforeseeable use, and/or improper use of ChatGPT." It pointed to terms of use provisions barring under-eighteens from using the product without parental consent, prohibiting use for suicide or self-harm, and forbidding circumvention of protective measures. It argued that the full chat history showed recurring suicidal ideation predating any use of the product, that ChatGPT directed the user to crisis resources and trusted people more than a hundred times, and that he circumvented safety features by framing queries as fiction. The filing drew heavy public criticism.

Note what is not in it. There is no motion to compel arbitration, and there is no reported ruling in this proceeding on the two defences the industry is actually watching. Section 230 of the Communications Decency Act immunises interactive computer services for information provided by another information content provider; generated output arguably has no other provider, which is why the defence has not carried the day in the chatbot cases so far. The First Amendment defence fared no better in the closest analogue: in Garcia v. Character Technologies, Inc., No. 6:24-cv-01903 (M.D. Fla.), Judge Anne Conway declined in May 2025 to dismiss on First Amendment grounds, expressing scepticism that large language model output is "speech" for constitutional purposes and allowing product liability, negligence and consumer protection claims to proceed. That case has since reportedly resolved as part of a package of settlements, which means the ruling stands as persuasive authority with no appellate test.

What it means for claimants' counsel, defence counsel and clause drafters

For claimants' counsel. The forum question is live and it is not symmetrical. A new California case joins JCCP 5431 by an add-on request under California Rules of Court rule 3.544, submitted to the coordination trial judge under section 404.4 with the showings required by rules 3.520 through 3.523; opponents have ten days to serve a notice of opposition and fifteen days after that to file a memorandum, and if no opposition is filed the coordination trial judge may grant or deny without a hearing. That is a fast, cheap route into an organised proceeding with appointed leadership and, on the face of CMO No. 1, no surrender of settlement authority. Filing federally buys none of that today, because there is no MDL to join. The countervailing risk is a Statute of Limitations clock that does not wait for coordination: California's two-year personal injury period under Code of Civil Procedure section 335.1 runs, and the coordination machinery does not toll it.

For defence counsel. The arbitration question should be decided, not deferred. Uhl is a Ninth Circuit decision about a consumer platform that waited eleven months, and the reasoning is portable to any defendant that litigates the merits while holding a clause in reserve. Where a defendant intends to invoke a clause against signatory plaintiffs, the move belongs before the merits pleading, with a stay sought for the rest. Where it does not intend to invoke it, saying so early removes a hostage that plaintiffs will otherwise use to characterise every scheduling dispute.

For clause drafters. These cases are a stress test of what a consumer Arbitration Clause can and cannot reach. A clause that binds "heirs" does not, in California, reach an heir's independent wrongful death claim, and the drafting cannot fix that by adjective; a genuinely independent claimant has not agreed to anything. A 30-day opt-out helps on procedural Unconscionability and does nothing on scope. If the exposure a product creates is catastrophic personal injury rather than fee disputes, the clause is not the control the risk register thinks it is, and the design, warning and escalation record is.

For funders and firm economics. A coordinated proceeding without an inventory, without a Census Registry and without a Common Benefit Fund order behaves very differently from a federal mass tort. Common benefit work is being done by four co-leads and a steering committee with no publicly ordered holdback to pay for it. That is a question the proceeding will have to answer, and it is one to watch on the schedule that comes out of today's conference.

Frequently asked questions

How many ChatGPT product liability lawsuits are pending?

Roughly twenty-three actions are coordinated in JCCP 5431 and the total against OpenAI across state and federal courts exceeds twenty, including a wave of about thirty complaints filed in California federal court in early September 2026 arising from the February 2026 Tumbler Ridge school shooting; the exact number moves weekly as cases are filed and added on.

Is JCCP 5431 the same thing as an MDL?

No — a Judicial Council Coordination Proceeding is California's state-court equivalent, created under Code of Civil Procedure section 404, and it reaches only California state actions, whereas an MDL is created by the Judicial Panel on Multidistrict Litigation (JPML) under 28 U.S.C. § 1407 and reaches federal actions nationwide.

Who is the judge in the ChatGPT product liability JCCP?

Judge Ethan P. Schulman of the San Francisco County Superior Court, sitting in Department 304 as coordination trial judge.

Has OpenAI tried to send these cases to arbitration?

Not on the public record: despite a binding individual arbitration clause and class action waiver in its consumer Terms of Use, OpenAI has not moved to compel arbitration in the coordinated cases, and its answer in Raine pleaded misuse and terms-of-use breach rather than a demand for arbitration.

Does Section 230 or the First Amendment protect ChatGPT's outputs?

Neither defence has succeeded in a chatbot injury case to date — Section 230 protects information provided by another information content provider, which generated output arguably is not, and in Garcia v. Character Technologies the court declined to dismiss on First Amendment grounds, doubting that large language model output is constitutionally protected speech — but no appellate court has yet ruled, so both remain open.

What is the next date in the ChatGPT product liability JCCP?

Today, 23 September 2026, is the further case management conference in Department 304; any subsequent schedule for master pleadings, demurrers and discovery is expected to come out of that conference.