B.C. and SD59 file in N.D. Cal. on 21 September

On 21 September 2026, British Columbia Attorney General Niki Sharma announced that the Province was filing a California lawsuit against OpenAI over alleged failure to notify law enforcement of threats on its platform before the 10 February 2026 mass shooting at Tumbler Ridge Secondary School. The release says the Province is working with the board of education for School District No. 59 (Peace River South). Sharma also says she wrote to federal ministers Fraser, Solomon, and Miller with proposals to amend the Criminal Code of Canada on AI accountability.

The same day, counsel filed a 39-page federal complaint in the U.S. District Court for the Northern District of California, San Francisco Division. The caption is His Majesty the King in Right of the Province of British Columbia and Board of Education of School District No. 59 (Peace River South) v. Samuel Altman et al. The complaint states Case No. 3:26-cv-10743. The ECF header on the hosted PDF also shows Case 4:26-cv-10743-KAW, Document 1, Filed 09/21/26. That division stamp variance is unresolved on the face of the PDF; we cite the caption's Case No. 3:26-cv-10743.

Plaintiffs are the Crown (B.C.) and SD59. Named defendants are Samuel Altman; OpenAI Foundation (f/k/a OpenAI, Inc.); OpenAI OpCo, LLC; OpenAI Holdings, LLC; and OpenAI Group PBC. U.S. counsel of record are Lesley E. Weaver, Anne K. Davis, and Joshua D. Samra of Stranch, Jennings & Garvey, PLLC. Canadian counsel listed are Reidar Mogerman, K.C., and Rebecca Coad of CFM Lawyers LLP. The firm confirms the same caption and case number.

This investigation treats the AG statement and the filed complaint as the primary record. ChatGPT safety-review and warning narratives below are plaintiff allegations in the complaint unless separately labeled as quoted admissions the complaint attributes to OpenAI.

39-page complaint on Tumbler Ridge

Incident as pleaded. The complaint alleges that on the morning of 10 February 2026, eighteen-year-old Jesse Van Rootselaar of Tumbler Ridge killed their mother and eleven-year-old half-brother at the family home, then drove to Tumbler Ridge Secondary School (TRSS), operated by SD59. At the school, the complaint alleges Van Rootselaar killed five students and an education assistant, wounded two students who were transported for care, and died by suicide. A later paragraph states six people were killed inside the school in addition to the shooter: a 39-year-old female education assistant and five children ages 12 and 13. Approximately twenty-five injured sought medical care after evacuation, per the complaint. Roughly 160 students, teachers, and staff were trapped during the assault and recovery, the complaint says. The AG release does not enumerate victims; we use the complaint's figures only and do not reconcile secondary wire counts here.

Core liability theory (allegations). Plaintiffs allege OpenAI's then-internal procedures included automated review of ChatGPT content that could escalate to a human safety team. In June 2025, the complaint alleges, automated monitoring identified Van Rootselaar's account and routed it to that human review team. The team allegedly concluded there was a credible, specific risk of harm to others and recommended referral to the Royal Canadian Mounted Police (RCMP). The complaint alleges OpenAI leadership rejected that recommendation, deactivated the flagged account privately, notified no one outside the company, and that Van Rootselaar continued ChatGPT engagement through a second account. Plaintiffs allege these facts became public through anonymous OpenAI whistleblowers who reported to The Wall Street Journal. The complaint further alleges that one telephone call to the RCMP could have prevented the tragedy, and that the RCMP contracts with the Province to police Tumbler Ridge.

Quoted OpenAI statements in the complaint. The complaint quotes a 23 April 2026 letter attributed to Samuel Altman to the Tumbler Ridge community: "I am deeply sorry that we did not alert law enforcement to the account that was banned in June." The complaint treats that line as an admission that OpenAI identified the risk and failed to act. It also alleges OpenAI has acknowledged that under escalation guidelines adopted after June 2025, the shooter's account "would have been referred to law enforcement." Those quotations appear in the complaint citing secondary press. For this draft we report them as complaint content. We have not independently verified the letter originals.

Eight counts. The complaint demands a jury trial and lists:

  1. Negligence (Failure to Warn Law Enforcement)
  2. Negligent Entrustment
  3. Aiding and Abetting a Mass Shooting
  4. Negligence (Failure to Warn)
  5. Negligent Undertaking
  6. Negligence (Design Defect)
  7. Strict Product Liability (Design Defect)
  8. Strict Product Liability (Failure to Warn)

All eight are pleaded against all defendants. Diversity jurisdiction is asserted with amount in controversy exceeding $75,000.

Relief sought. Plaintiffs seek compensatory damages for emergency response, health care, victim services, educational continuity, facility remediation and demolition, and construction of a replacement school and community wellness center, in amounts to be proven at trial; punitive damages, restitution, and disgorgement as permitted; costs and interest; attorneys' fees; and declaratory relief. Injunctive asks include, among other items: safeguards that refuse or de-escalate violence-intent conversations; automated detection plus human-review escalation; a policy of referring imminent or credible threats of serious physical harm to law enforcement; warnings about safety degradation in extended multi-turn chats; durable account-level enforcement against re-registration after violent-misuse bans; a "red-line" principle against facilitating violence; auditable safety controls; and quarterly compliance audits by an independent monitor. The complaint also proposes prompt production of Van Rootselaar's chat logs with leave to amend.

Provincial cost frame. The complaint alleges the Province declared a provincial state of emergency on 11 February 2026 and mobilized extraordinary, non-routine resources. SD59 owns facilities the complaint says were damaged and rendered unusable and had to stand up temporary education capacity.

AG push and counsel demand for damages

The AG release frames the filing as part of a broader push for answers, accountability, and stronger national AI safeguards. Sharma says OpenAI "has not taken meaningful steps to address the concerns raised by survivors, families and the Tumbler Ridge community," and still has an opportunity to strengthen safeguards and improve transparency. The release links community supports and notes prior family legal filings related to Tumbler Ridge.

Counsel's firm notice restates the demand for damages and injunctive safeguards and quotes Sharma on technology-company responsibilities when credible threats of serious violence are known. Secondary coverage, including Ars Technica (which hosts the complaint PDF) and CBC, tracks the same filing day. Those wires are color only; they are not our primary gate.

Separately, the complaint itself alleges earlier public pressure (Premier David Eby and Tumbler Ridge Mayor Darryl Krakowka in April 2026) and family suits. Those are distinct from this Crown/SD59 action and should not be conflated without their own dockets.

Treat as allegations and map duty-to-warn

For operators on policy, legal, and school-safety desks, the Monday read of this record is procedural:

  • Treat the complaint as allegations. No court has adjudicated the June 2025 review, the RCMP-referral claim, the second-account path, or product-defect theories.
  • Map duty-to-warn exposure. If your organization runs consumer or education-facing chat systems with violence-escalation queues, document who can override a human reviewer's law-enforcement referral, what written criteria apply, and whether deactivation without external notice is an authorized outcome.
  • Separate product bans from identity persistence. The complaint's injunction ask on re-registration after violent-misuse bans is a concrete control question: same identity, devices, or payment method. Ask vendors for enforceable account-level durability in addition to content filters.
  • Preserve vendor communications. Provinces, districts, and campus safety leads that have asked labs for post-incident safeguards should keep the ask trail. The AG release says concrete action is still expected.
  • Watch Canadian federal Criminal Code proposals. Sharma's letters to ministers Fraser, Solomon, and Miller are policy signals. They are not enacted statute. Track whether a referral pathway or accountability amendment is introduced.
  • Do not use this filing as proof of ChatGPT causation. Liability theories here are contested pleadings in N.D. Cal.

Whistleblower claims and unproven outcomes

  • That OpenAI's safety team in fact made an RCMP-referral recommendation, or that leadership rejected it. Those are complaint allegations attributed in part to whistleblowers.
  • That a warning would have stopped the attack. The complaint asserts that theory; it is unproven.
  • That ChatGPT "incited" the shooter or that design/defect counts will succeed. Counts three through eight are claims for trial. They are not findings.
  • Victim counts from secondary wires that differ from the complaint. We stay with the complaint's school and home figures and note the AG release is silent on counts.
  • OpenAI's full responsive pleading or any PACER docket sheet beyond the hosted Document 1 PDF. We did not retrieve a defendant answer for this draft.
  • Graphic crime-scene detail. The operational facts above are limited to what the complaint and AG statement put in the public record for liability and relief.

PACER docket, OpenAI answer, and chat logs

  • A PACER docket sheet confirming assignment, judge initials, and any subsequent ECF activity beyond Document 1.
  • OpenAI's answer, motion practice, or a company primary statement responding to this Crown/SD59 complaint specifically.
  • Production of the chat logs the plaintiffs say they will seek; the complaint flags leave to amend after production.
  • Independent verification of the April 2026 Altman letter text beyond the complaint's secondary-press citations.
  • Text of Sharma's letters to federal ministers Fraser, Solomon, and Miller.
  • Any court order on the injunctive package or on chat-log production deadlines.
  • Reconciliation, if any, between complaint victim enumerations and other official RCMP tallies we have not re-fetched for this draft.

Monday action: Policy counsel and school-safety leads should brief boards on the caption, eight counts, and injunction checklist as alleged vendor duties under this filing; open a vendor questionnaire on law-enforcement referral override authority and ban durability; and calendar N.D. Cal. Case No. 3:26-cv-10743 for the answer deadline once the docket is confirmed. Do not treat the complaint narrative as adjudicated fact.