Mass tort economics usually rest on reference points. A firm signing a talc case or an AFFF case can look at prior verdicts, settlement grids, Daubert history, and MDL structure, and price the inventory accordingly. The AI wrongful death docket has none of that. There is no verdict, no settlement program, no admissibility ruling, and no consolidated proceeding. There is a body of complaints, one answer, one instructive ruling from an adjacent case, and a filing pace that keeps accelerating anyway.
That combination, fast-growing inventory against unpriced risk, is the most interesting business story in plaintiff-side law right now. It is worth taking apart.
The docket, briefly
Raine v. OpenAI, filed August 26, 2025 in San Francisco County Superior Court, is the foundation. The parents of a 16-year-old who died by suicide sued OpenAI and Sam Altman personally, pleading seven causes of action anchored on California strict products liability. An amended complaint in October 2025 sharpened the theory from rushed-to-market negligence to intentional misconduct, alleging the deliberate removal of self-harm safeguards and opening the path to punitive damages through the survival action. OpenAI answered in November, denying responsibility and arguing the harm resulted from misuse in violation of its terms of use.
Around that anchor, the docket widened in waves. November 2025 brought a coordinated set of additional suits in California courts alleging deaths and psychological harm tied to extended ChatGPT use. Then, in a six-week stretch this spring: seven federal suits over the Tumbler Ridge school shooting filed in the Northern District of California in late April, the FSU shooting case filed in federal court in Florida in May with a novel negligent entrustment count, and the Scott overdose case filed in California state court days later, alleging the model affirmatively misinformed a user about a fatal drug combination. Different harms, same architecture: ChatGPT as a product, safeguards as the defect, internal data as the proof.
The only merits-adjacent signal in the entire space remains Garcia v. Character Technologies, where a Florida federal court in May 2025 allowed product liability claims against a chatbot maker to proceed past dismissal and declined to resolve the First Amendment defense at the pleading stage. One survival of one motion in one adjacent case is, at present, the closest thing this docket has to a comparable.
Who is building the inventory
This is not, so far, a broad-intake docket. A small group of repeat players dominates the filings: Edelson PC, which brought the Tumbler Ridge cases and appears in the Raine matter; the Tech Justice Law Project, whose involvement runs through the docket's most consequential pleadings; and the Social Media Victims Law Center, which carried its client base and theory-building from the social media addiction litigation directly into the AI cases. The continuity matters. The firms that spent the last three years litigating design-defect theories against platform defendants arrived at the AI docket with the doctrinal scaffolding, the expert relationships, and the client pipelines already in place.
For everyone else, the strategic question is the familiar one from every emerging tort: enter now, while the law is unsettled and the signing criteria are still being invented, or wait for the dispositive rulings and pay for the delay in referral fees. The difference here is the steepness of the curve. In most emerging torts the science is the gating uncertainty. In this one it is the law itself.
Signing criteria without a diagnosis code
Conventional mass tort intake screens on two artifacts: proof of product use and a qualifying diagnosis. The AI cases compress both into a single artifact that no prior docket has had, the complete conversation history between the product and the person harmed.
That changes the intake calculus in several ways. The chat log is simultaneously the product identification, the exposure history, and the core causation evidence, and it exists in the defendant's possession in a form the plaintiff can usually also export. Screening a case means reading it. The strongest files in the current docket share a recognizable pattern: extended engagement over months, escalating dependence, explicit warning signs the system demonstrably registered, and output that a jury could read as participation rather than passivity. The weakest files are short, ambiguous, or missing the account access needed to reconstruct the record.
Preservation is the immediate practice point. Account deletion, conversation deletion, and model deprecation all threaten the central evidence, and the families who become clients are often the same people most likely to delete an account in grief. Firms working this space have learned to treat data export and litigation hold notices as day-one tasks, ahead of nearly everything else.
Timing carries a damages dimension that has already shaped the docket once. California's Senate Bill 447 window, which allowed survival-action plaintiffs to recover for a decedent's pre-death pain and suffering in cases filed before January 1, 2026, helps explain the density of late-2025 filings. Cases filed after the window closed lose that component unless the legislature extends it, a quiet but real difference in case value that intake teams now have to price by filing date.
The discovery playbook Raine wrote
The Raine amended complaint did more than add a punitive damages theory. It established what discovery in these cases is for. The complaint's evidentiary core is OpenAI's own moderation data, the allegation that the company's systems flagged hundreds of messages for self-harm content without intervention. Detection worked; response did not. That gap between what the system saw and what it did is the structural argument, and it points discovery at a specific document set: moderation logs, safety policy documents and their revision history, internal debates over guardrail changes, and the launch-decision record for each model version.
Nearly every complaint filed since asks for some version of the same record. The Tumbler Ridge cases allege internal flagging of the eventual shooter months before the event. The Scott case patterns on the removed-safeguards theory directly. For defendants, this convergence means the same internal documents are now relevant across multiple proceedings in multiple courts, which is precisely the condition that historically produces coordination pressure, and eventually, in dockets that survive their early motions, settlement pressure.
The consolidation math
There is no MDL and no immediate prospect of one. The docket is split between California state court, where Raine, Scott, and much of the November wave sit, and federal courts in California and Florida. If state filings keep accumulating, the natural mechanism is a Judicial Council coordination proceeding, the same JCCP structure currently organizing the social media addiction litigation. The federal cases are too few and too factually distinct, for now, to support an MDL petition that the Panel would take seriously.
The practical consequence is that the early rulings will come retail, court by court, with all the inconsistency that implies. That is a risk and an opportunity in equal measure. A favorable ruling in one court does not bind the next, but it builds the persuasive record, and the firms steering the docket are visibly sequencing their arguments with that in mind.
Pricing the option
Strip away the technology and the inventory question is familiar: what is a contingency case worth when liability law is unresolved? Three features distinguish this docket's version of the problem.
First, the defendant can fund any defense indefinitely. OpenAI's most recent reported valuation puts it among the most valuable private companies in the world, and unlike the asbestos or talc defendants of prior generations, it faces no solvency pressure to resolve anything. Settlement, if it comes, will be a reputational and regulatory calculation, not a balance-sheet one.
Second, the insurance picture is genuinely unclear. Whether and how existing commercial liability programs respond to AI output claims is an open question being worked out in parallel, and the answer shapes both settlement capacity and the willingness of carriers to force early resolution. Firms valuing inventory against an uninsured or under-insured liability are making a different bet than the one mass tort economics usually assumes.
Third, the repricing event has a date range. Dispositive motion rulings across the docket are expected in the second half of 2026. If the product framing survives in even one or two courts, the inventory built over the past year converts from speculative to conventional, the referral market floods, and the firms that signed early own the cases everyone else now wants. If the early rulings go the other way, the docket does not die, the theories are too varied and the courts too many for a single ruling to end it, but the option premium evaporates and the next round of filings gets much more selective.
Either way, the present moment is the cheap part of the curve. That is what the filing pace is telling you. The firms building this inventory are not confused about the uncertainty. They are pricing it, and they have concluded that the cost of being early is lower than the cost of being late.
The claims described in this article are allegations by the parties and have not been adjudicated. Case status is current as of publication; the consumer-facing case tracker at Lawsuit Informer is updated as the docket moves.