The seven complaints filed Wednesday in the Northern District of California by the families of the twelve people killed in the Tumbler Ridge mass shooting were lodged on PACER's public docket Friday at 4:48 p.m. Pacific. The texts confirm the central allegation Edelson PC made in its press conference Tuesday: that on June 14, 2025, an internal review team of twelve OpenAI engineers received an automated flag against the gunman's account, recommended a referral to the Royal Canadian Mounted Police, and was overruled by management, which deactivated the account and "kept what they had seen to themselves." [1] The paper's Friday account of the lawyers' framing noted the docket text was pending. The texts are now public.

The complaint identifies the twelve engineers by job title rather than by name — "Trust and Safety Reviewer," "Senior Trust and Safety Engineer," "Director of Trust and Safety Operations" — with the count of each role on the team. The pleading does not allege that the management override was authorized by Sam Altman by name, but it does allege that "C-suite executives" approved the deactivation-only response and that the decision is documented in an internal Slack channel and a Google Docs file referenced as "BL-2025-06-14-Triage-Decision.docx." Both digital artifacts are the subject of preservation orders the plaintiffs filed alongside the complaints. [2]

The relief sought goes beyond damages. The seven complaints, drafted as parallel filings, ask the court to order OpenAI to (a) implement a mandatory law-enforcement-referral protocol for accounts flagged with violence-planning indicators, (b) publish quarterly transparency reports on flag-to-referral conversion rates, (c) submit to third-party audit of its trust-and-safety operations for three years, and (d) compensate the families of the twelve victims. The damages calculation is unspecified; the structural-overhaul relief is the lead. The combined demand of "in excess of $1 billion" applies across all seven cases. [3]

The June 14, 2025 date is the document's load-bearing claim. The Tumbler Ridge attack occurred on July 18, 2025 — thirty-four days after the alleged flag. The complaint describes the gunman's ChatGPT use during that window: 192 sessions averaging forty-three minutes each, with prompts the pleading characterizes as "operational planning, weapons selection, and target reconnaissance." The flag, according to the complaint, classified the conduct as "gun violence activity and planning" under OpenAI's own internal taxonomy. The deactivation occurred forty-eight hours after the flag and was implemented without notification to law enforcement in either Canada or the United States. The thirty-two days between deactivation and attack are, in the plaintiffs' framing, the window during which a referral could have prevented the deaths. [4]

The legal community's read on the document is that the structural-overhaul relief is the actual stake. A wrongful-death tort against a foundation-model lab on a theory of negligent failure to warn is a novel-but-recognizable cause of action; courts have struggled with similar theories in cases involving social-media platforms, search engines, and content recommendation systems. The structural relief, by contrast, would require the court to impose ongoing operational duties on a private company in the absence of a regulatory framework. Federal courts have ordered structural relief in civil-rights cases against police departments and prison systems; against private technology companies, the precedent is thinner. [5]

The Cerebras roadshow opens Monday at the Bloomberg roadshow venue in midtown Manhattan with a $4 billion target and a $40 billion valuation indication. The roadshow window is also the discovery window in the Tumbler Ridge cases — the first scheduling order is expected from N.D. Cal. before May 15. Cerebras's S-1 filing, which references "ongoing product-liability litigation involving Cerebras's largest commercial customer" in its risk-factors section, is the institutional artifact that connects the two windows. The customer the language references is OpenAI, which is the subject of the Tumbler Ridge complaints, the SDNY counterparty under Florida AG subpoena, and the IPO-window-vs-discovery-window collision the paper has tracked since April 22. [6]

OpenAI filed no public response to the complaints Friday or Saturday. The company's standard practice in product-liability matters is to respond at the answer deadline, which under Federal Rule 12(a)(1)(A) is twenty-one days from service. The complaints were served Wednesday afternoon. The answer is due May 20. The first scheduling order is likely to come earlier, possibly with an OpenAI motion to consolidate the seven cases under one master docket — a procedural move that would compress the discovery timeline and is itself a litigation choice. The companies that have made the choice in similar product-liability multidistrict litigation have generally been the companies that intended to settle. [7]

-- THEO KAPLAN, San Francisco