British Columbia Sues OpenAI: The Tumbler Ridge Claims and the Duty to Warn

British Columbia sues OpenAI: what was filed

The claim is a 39-page complaint in 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. Altman et al., Case No. 3:26-cv-10743, filed in the San Francisco Division of the US District Court for the Northern District of California. The plaintiffs are the provincial Crown, represented by Attorney General Niki Sharma, and the board that runs public schools in the Peace River South region, including the school where the shooting took place.

The defendants are Samuel Altman individually and four OpenAI entities: OpenAI Foundation (formerly OpenAI, Inc.), OpenAI OpCo, LLC, OpenAI Holdings, LLC and OpenAI Group PBC, the public benefit corporation the complaint says was formed on 28 October 2025 in OpenAI's restructuring. The plaintiffs demand a jury trial. They are represented by Stranch, Jennings & Garvey in Oakland, with CFM Lawyers in Vancouver acting as Canadian counsel on any questions of Canadian law.

Announcing the filing, the Attorney General said in a government statement that the province was suing "for its failure to notify law enforcement of threats made on its platform prior to the mass shooting". CBC News reported that Sharma also described a second limb, a claim that OpenAI designed an unsafe product. For Canadian public bodies and the advisers in Canada who act for them, the case is an unusual step: a province litigating its own losses in a foreign court against a technology company and its chief executive.

What happened in Tumbler Ridge

On 10 February 2026 an 18-year-old resident of Tumbler Ridge, a community of about 2,400 people in northeastern British Columbia, killed eight people before dying of a self-inflicted wound. According to police accounts summarised in the complaint and reported by Reuters, the shooter, identified by police as Jesse Van Rootselaar, first killed their mother and 11-year-old half-brother at the family home, then killed five students aged 12 and 13 and an education assistant at Tumbler Ridge Secondary School. Two students were airlifted to hospital in Vancouver and about 25 other people sought treatment locally.

The complaint describes it as the deadliest school shooting in Canada in more than three decades. The province declared a state of emergency on 11 February. The school has since been closed and, following community consultation, demolished; demolition began in August 2026. A separate criminal investigation by the RCMP remains open, and CBC reported that British Columbia's chief coroner has directed an inquest into what led up to the attack.

What the complaint alleges about the June 2025 review

The core allegation is that OpenAI knew of a specific risk months before the attack and chose not to share it. The complaint alleges that in June 2025 OpenAI's automated monitoring identified substantial activity on the shooter's account involving gun-violence scenarios and routed it to the human team OpenAI maintains to review users who may be planning to harm others. That team, it says, concluded there was a credible and specific risk of harm to others and recommended a referral to the Royal Canadian Mounted Police.

OpenAI leadership, the province alleges, rejected the recommendation, taking the position that the matter did not meet a "higher threshold" for "credible and imminent" threat reporting. The account was deactivated and no one outside the company was told. The complaint states that roughly a dozen employees reviewed the flagged conversations, and it relies on whistleblower accounts first reported by The Wall Street Journal in February 2026.

Four further strands of the pleading are likely to matter as the case develops:

• The second account. The province alleges that deactivating one account did not stop the shooter, who continued to use ChatGPT through a second account until the day of the attack, and that OpenAI's help pages explain to deactivated users how to regain access.
• Identifying data. It argues that OpenAI already held the name, email, IP address and approximate location data a referral would have needed, which it says is hard to reconcile with a privacy rationale for staying silent.
• Prior incidents. It alleges OpenAI was on notice from earlier attacks, including the April 2025 Florida State University shooting, the January 2025 Las Vegas Cybertruck explosion and a May 2025 school stabbing in Finland.
• Design choices. It alleges that GPT-4o was tuned for engagement, with memory on by default and a Model Spec that told the model to "assume best intentions" and not to ask users to clarify their intent.

The complaint also says that OpenAI's own later guidance concedes the point: under escalation guidelines adopted after June 2025, it quotes OpenAI as saying, the account would have been referred to law enforcement. The chat logs themselves have not been disclosed. Sharma told reporters she has not read them and that OpenAI refused the province's request to see them.

The eight causes of action

The complaint pleads eight counts against all defendants, mixing classic negligence theories with California strict product liability and an intentional-tort count that will draw most of the attention. Each count is a matter for proof; none has yet been answered.

Count	Cause of action	The province's theory in brief
1	Negligence (failure to warn law enforcement)	OpenAI's reviewers identified a credible risk and the company owed, and breached, a duty to tell the RCMP
2	Negligent entrustment	OpenAI kept supplying a powerful product to a user it had flagged as dangerous
3	Aiding and abetting a mass shooting	With actual knowledge of the risk, OpenAI gave substantial assistance through ChatGPT's configuration; punitive damages sought
4	Negligence (failure to warn)	OpenAI failed to warn of known or foreseeable risks of ChatGPT
5	Negligent undertaking	Having undertaken to detect and refer dangerous users, OpenAI performed that task carelessly
6	Negligence (design defect)	ChatGPT was negligently designed to prolong engagement and permit violent content
7	Strict product liability (design defect)	ChatGPT is a product under California law and was defectively designed
8	Strict product liability (failure to warn)	The product lacked adequate warnings about its capacity to facilitate harm to others

The aiding and abetting count is the most aggressive. It alleges that OpenAI acted with knowledge that ChatGPT was being used to plan real-world violence and made "a conscious and intentional decision" to keep supplying the shooter. Aiding and abetting a tort ordinarily requires knowledge of the primary wrong and substantial assistance, a high bar where the assistance alleged is the continued availability of a general-purpose tool. Litigators in product liability will watch whether the court treats the negligence counts as the real centre of gravity.

What the Province and SD59 are asking the court for

The province and the school board want their money back and a court order to change OpenAI's conduct. The complaint says the costs claimed are "extraordinary, non-recurring" public expenditures rather than ordinary government spending, a framing that appears designed to meet the argument, familiar in public-entity litigation in the US, that governments cannot recover the routine cost of public services from a wrongdoer.

Relief sought	What the complaint asks for
Compensatory damages	Past and future costs of policing and emergency response, health care, victim services and trauma-informed mental-health care, educational continuity and replacement staff, remediation and demolition, and a replacement school and wellness centre
Punitive damages, restitution and disgorgement	In an amount permitted by law
Injunction: refusal and de-escalation	Safeguards that reliably refuse, end or de-escalate conversations where a user expresses intent to harm others
Injunction: detection and referral	Automated detection with human-review escalation, and a policy of referring users who present an imminent or credible threat to law enforcement
Injunction: warnings and framing	Prominent warnings that safety features degrade in long conversations, and no fictional or "research" framings to unlock violent content
Injunction: repeat users	Durable account-level enforcement so a banned user cannot promptly re-register on the same identity, devices or payment method
Injunction: oversight	An enforced anti-violence red line, auditable controls and quarterly audits by an independent monitor
Declaration	That the product was defectively designed and lacked adequate warnings

No figure is pleaded; damages are "to be proven at trial". CBC reported that Sharma said the province is seeking reimbursement for building a new Tumbler Ridge school, and that the province's claim for damages is "quite different" from the families' claims because it concerns the province's own losses.

What OpenAI has said

OpenAI has not yet filed a response to the provincial complaint. In a statement to Reuters, spokesperson Drew Pusateri called the shooting "an unspeakable tragedy" and said: "OpenAI remains committed to working collaboratively with government and law enforcement officials, and continuing to advance our ongoing safety work." Reuters also reported that OpenAI, which has denied the claims, says it trains its models to refuse requests that could "meaningfully enable violence" and notifies law enforcement when conversations suggest "an imminent and credible risk of harm to others".

The company's earlier public statements are central to the pleading. In a letter of 26 February 2026 to Canada's AI minister, Evan Solomon, OpenAI's vice president of global policy, Ann O'Leary, wrote that when the account was banned in June 2025 the company "did not identify credible and imminent planning that met our threshold" for a referral. The same letter said that under its enhanced protocol OpenAI "would refer the account banned in June 2025 to law enforcement if it were discovered today", and that it discovered the second account only after the shooter's name was released, then shared it with police. It committed to direct points of contact with Canadian law enforcement and stronger detection of repeat violators.

On 23 April Sam Altman wrote to the community. CBC reported the letter, which OpenAI confirmed as authentic: "I am deeply sorry that we did not alert law enforcement to the account that was banned in June." The complaint calls that sentence "an admission that OpenAI identified the risk and failed to act on it", and, as Business Insider reported, alleges that neither Altman nor O'Leary followed through on promised reforms. In the private cases OpenAI has disputed several allegations; its chief strategy officer Jason Kwon told NPR it was "absolutely false" to say that its chief global affairs officer, Chris Lehane, was involved in the original referral decision.

Why California: jurisdiction and the forum fight

The province chose OpenAI's home court. The complaint invokes federal diversity jurisdiction under 28 U.S.C. § 1332(a)(2) and (a)(4), the provisions that cover claims between citizens of a US state and foreign parties, including a foreign state as plaintiff, and says the design, the chats and, on information and belief, the decision not to warn the RCMP all happened in California. Sharma told reporters, as Black Press Media reported, that filing there gives the province "different ability and rules to go after the company" and supports the families who had already sued in California.

That choice collides with OpenAI's strategy in the private cases. On 17 September OpenAI moved to dismiss them on the ground of forum non conveniens, arguing that British Columbia is the more appropriate forum. CBC quoted the motion: "California is not the most appropriate forum for claims by British Columbians attacked by a British Columbian in British Columbia." OpenAI says most evidence and witnesses are in the province, that the case will require examination of Canadian law enforcement, firearms regulation and mental-health institutions, and that British Columbia has "a uniquely strong sovereign interest in having these actions resolved at home". It also argues that Altman "was not personally involved" in the June 2025 review.

The test comes from Piper Aircraft Co. v. Reyno (1981), under which a court asks whether an adequate alternative forum exists and weighs private and public interest factors, giving less deference to a foreign plaintiff's choice of a US forum. The provincial claim complicates that analysis. It is awkward for a defendant to argue that British Columbia's sovereign interest requires the case to be heard at home when the sovereign itself has chosen California. Cross-border litigators and California counsel will watch whether the court considers the motions against the backdrop of the new filing; the private motions are listed for hearing on 5 November.

How the provincial claim differs from the 37 private suits

The provincial action is the 38th Tumbler Ridge case against OpenAI in the same court. Families of victims filed seven suits in April, and 30 more were filed on 2 September on behalf of students, teachers and others present at the school, according to NPR. Those later complaints allege that the referral recommendation was overruled by the company's global affairs team, an allegation OpenAI denies.

Feature	Private suits (37)	Provincial and SD59 claim
Plaintiffs	Families of victims, survivors, students, teachers and staff	The provincial Crown and the school board
Losses claimed	Personal injury, wrongful death and emotional harm	Public expenditure, the destroyed school and ongoing recovery costs
Defendants	OpenAI and Sam Altman	Sam Altman and four named OpenAI entities
Injunctive relief	Not the main focus of reporting	An eight-part injunction with an independent monitor
Forum status	OpenAI motions to dismiss for forum non conveniens, hearing 5 November	Newly filed; no response yet

Mass-tort dockets of this kind tend to be coordinated before a single judge, as happened in the consolidation of more than 3,000 sexual assault claims against Uber. Whether the provincial case is managed alongside the private suits will shape its pace. CP24 reported that Premier David Eby accused OpenAI of walking away from mediation before the second wave of suits was filed.

Is there a duty to warn? Tarasoff and its limits

The central legal question is whether an AI provider that learns a user may be dangerous owes a duty to third parties to warn the police. US tort law generally imposes no duty to protect others from a third party's criminal acts unless there is a special relationship or the defendant has created or increased the risk. The best-known exception is Tarasoff v. Regents of the University of California (1976), in which the California Supreme Court held that a therapist who determines a patient presents a serious danger of violence to another must use reasonable care to protect the intended victim. "The protective privilege ends where the public peril begins," the court said.

The province's pleading does not rest on a therapist analogy alone. It argues that OpenAI created the duty itself: the company published a process under which trained reviewers could refer imminent threats to police, its reviewers applied that process and recommended a referral, and leadership overrode them. That is the logic of the negligent undertaking count, which does not depend on a free-standing duty to the world but on the claim that a party who voluntarily undertakes a protective task must perform it with care.

Several defences are foreseeable. OpenAI can argue that no identified victim was threatened, a feature of Tarasoff, that hindsight should not set the standard, and that a warning would not have prevented the attack. Its motion in the private cases already points to what it calls systemic failures by local institutions, and CBC quoted it raising whether police "acted reasonably in returning guns to the home". Causation, in other words, will be contested as hard as duty. For risk management advisers, the lesson is already visible: a published escalation policy may be treated as the measure of the duty.

Is ChatGPT a product? Section 230 and the design-defect theory

The complaint pleads carefully around Section 230 of the Communications Decency Act, which protects online services from being treated as the publisher of information provided by another content provider. It argues that ChatGPT generates its own output, so the claims "do not depend on treating OpenAI as the publisher or speaker of any third party's content", and that GPT-4o was a mass-distributed consumer product in the same way as a manufactured good.

That framing has some early support. In the Character.AI wrongful-death case in Florida, a federal judge in May 2025 allowed most product and negligence claims to proceed and said she was "not prepared" to hold that chatbot output was protected speech "at this stage", as the Associated Press reported. That is a first-instance ruling on a motion to dismiss, not an appellate holding, and it does not decide whether generative AI output is a product for strict liability purposes. The social media addiction cases, including the Tennessee jury trial against Meta and the TikTok settlement, show plaintiffs targeting platform design rather than user content to avoid Section 230.

Advisers in technology, media and telecoms should note the practical consequence: if model outputs are treated as the developer's own conduct, product design documents, model specifications and safety evaluations become core evidence. The allocation of responsibility across the AI supply chain is a question we also examined in our analysis of who is liable when law firms use AI.

Public bodies suing technology companies for public costs

The province frames this as part of a pattern. Sharma's statement lists tobacco and opioid manufacturers and vaping companies as earlier targets of British Columbia's litigation, and says the aim is that "people in B.C. are not left bearing the costs of corporate wrongdoing". What is new is the defendant: an AI developer, sued over a single violent event rather than a long-running public-health harm.

Public-body claims against technology companies have gathered pace in the US. A Kentucky school district settled its social media addiction claim for $27m, a New Mexico court ordered Meta to fund a $567m child mental-health remedy, and Texas has sued Netflix over allegedly addictive design. Reuters reported that Florida became the first US state to sue OpenAI over safety risks in June 2026. A school board joining a provincial government as co-plaintiff follows the same logic as the US school district cases: the institution that absorbed the cost sues in its own right.

The policy track in Canada

The lawsuit sits alongside a push for legislation. In her statement Sharma said she had written to federal ministers Fraser, Solomon and Miller "with clear policy ideas to amend the Criminal Code of Canada to ensure a pathway to human accountability for AI's actions", and that the case "highlights the urgent need for strong national safeguards for artificial intelligence technologies and online platforms". According to Black Press Media, she welcomed the federal government's proposed Social Media Act, Bill C-34, but said she does not think it goes far enough, adding: "We cannot allow these companies to self-regulate."

OpenAI's February letter to Minister Solomon already commits it to working with the Canadian government on referral criteria and "transparency regarding our reporting to law enforcement". Companies deploying AI in several jurisdictions will want to map those commitments against other regimes, including the obligations described in our guide to the EU AI Act implementation timeline, and to take cross-border advice on where a single global escalation policy will be judged.

What in-house counsel should take from the claim

Whatever the outcome, the complaint is a template for how future claims against AI and platform companies may be pleaded. It turns a company's own safety architecture into the case against it: the published referral process, the reviewers' recommendation, the executive override, the post-incident apology and the later statement that the account would now be referred. Legal and trust-and-safety teams can test their own position against it now:

• Escalation authority. Who can overrule a trust-and-safety recommendation to contact police, on what written criteria, and how is that decision recorded and reviewed?
• Published standards. Do public statements about when you refer threats match what your teams actually do? A gap between the two is the core of a negligent undertaking claim.
• Repeat users. Can a user removed for violent misuse return on a new account, and would you know? The second account is central to the province's case.
• Cross-border referrals. Do you have a route to police in each country where users sit, and a view on the local legal basis for disclosure? OpenAI committed to Canadian points of contact only after the attack.
• Post-incident statements. Apologies and admissions will be quoted back in pleadings. Coordinate communications with litigation counsel.
• Insurance. Check whether product liability, general liability and D&O programmes respond to claims framed as both product defect and personal liability of an executive; an insurance adviser can review wordings.

These are governance questions as much as litigation ones, and boards will expect answers before the next incident rather than after it. Specialist dispute resolution and litigation counsel can help a company stress-test its record against the theories in this complaint.

What happens next

OpenAI will respond to the provincial complaint in due course, and a motion to dismiss on forum, duty or Section 230 grounds would be the expected opening move. The complaint proposes an early deadline for production of the chat logs, after which the plaintiffs say they will seek to amend. In the private cases, the forum non conveniens motions are set for hearing on 5 November in San Francisco.

In British Columbia the RCMP investigation continues; a police spokesperson has said it has not ruled out charges, and the coroner's inquest is still to come. Those processes may produce the factual record the civil courts are currently arguing about. For now, British Columbia sues OpenAI on allegations drawn largely from whistleblower accounts and the company's own statements, and none of them has yet been tested in evidence.

Frequently asked questions

What does OpenAI have to do with Tumbler Ridge?

OpenAI has said its systems flagged the shooter's ChatGPT account in June 2025 and banned it, but that the activity did not meet its threshold for referral to police. British Columbia alleges that OpenAI's reviewers recommended alerting the RCMP, that leadership overruled them, and that the shooter kept using ChatGPT through a second account before the attack on 10 February 2026.

Why is British Columbia suing OpenAI?

The province says OpenAI failed to notify law enforcement of threats on its platform before the shooting and designed an unsafe product. It seeks to recover the public costs of the emergency response and recovery, including a replacement school, and a court order changing how OpenAI handles conversations that indicate a risk of violence.

Who are the parties to the case?

The plaintiffs are His Majesty the King in Right of the Province of British Columbia and the Board of Education of School District No. 59 (Peace River South). The defendants are Sam Altman and four OpenAI entities: OpenAI Foundation, OpenAI OpCo, OpenAI Holdings and OpenAI Group PBC. The case number is 3:26-cv-10743.

What legal claims does the complaint make?

Eight: negligence for failing to warn law enforcement, negligent entrustment, aiding and abetting a mass shooting, negligence for failure to warn, negligent undertaking, negligent design, and strict product liability for design defect and for failure to warn. The plaintiffs demand a jury trial and seek punitive damages.

Why was the lawsuit filed in California?

OpenAI is headquartered in San Francisco, and the province alleges the product design and the decision not to warn the RCMP took place there. The Attorney General said California gives the province different rules to pursue the company and allows it to support the families who had already sued in the same court.

How has OpenAI responded?

Spokesperson Drew Pusateri called the shooting an unspeakable tragedy and said OpenAI remains committed to working with government and law enforcement and advancing its safety work. OpenAI has not yet answered the provincial complaint. Sam Altman apologised in April for not alerting police to the account banned in June 2025.

How many lawsuits does OpenAI face over Tumbler Ridge?

The provincial claim is the 38th. Families of victims filed seven suits in April 2026 and 30 more were filed on 2 September on behalf of students, teachers and others at the school. OpenAI moved on 17 September to dismiss those private suits in favour of British Columbia's courts.

Does OpenAI report users to the police?

OpenAI says it notifies law enforcement when conversations suggest an imminent and credible risk of harm to others. In February 2026 it told Canada's AI minister it had made its referral criteria more flexible and would refer the June 2025 account if it were discovered today.

What happens next in the case?

OpenAI is expected to respond, possibly with a motion to dismiss. The plaintiffs have asked for early production of the chat logs and plan to amend once they have them. The forum motions in the private cases will be heard on 5 November, while the RCMP investigation and a coroner's inquest continue in British Columbia.

Sources

1. Complaint, His Majesty the King in Right of the Province of British Columbia v. Altman, Case No. 3:26-cv-10743 (N.D. Cal., filed 21 September 2026)
2. Government of British Columbia: Attorney General's statement on B.C. filing legal action against OpenAI (21 September 2026)
3. CBC News: B.C. government suing OpenAI for alleged role in Tumbler Ridge mass shooting (21 September 2026)
4. Reuters, via The Star: British Columbia sues OpenAI over Tumbler Ridge school shooting (21 September 2026)
5. Black Press Media, Victoria News: B.C. government files lawsuit against OpenAI in Tumbler Ridge shooting (21 September 2026)
6. CBC News: OpenAI asks California court to dismiss lawsuits over Tumbler Ridge shooting (17 September 2026)
7. CP24: OpenAI asks U.S. court to dismiss Tumbler Ridge mass shooting lawsuits (18 September 2026)
8. NPR: New lawsuits claim OpenAI execs put image ahead of safety in Canadian mass shooting (2 September 2026)
9. OpenAI: Letter from Ann M. O'Leary to Minister Evan Solomon (26 February 2026)
10. CBC News: OpenAI's Sam Altman writes apology to community of Tumbler Ridge (April 2026)
11. Business Insider, via Yahoo News: British Columbia files a lawsuit against OpenAI over mass shooting (22 September 2026)
12. Tarasoff v. Regents of the University of California, 17 Cal. 3d 425 (1976)
13. 47 U.S.C. § 230, Legal Information Institute
14. 28 U.S.C. § 1332, Legal Information Institute
15. Piper Aircraft Co. v. Reyno, 454 U.S. 235 (1981), Legal Information Institute
16. Associated Press, via Insurance Journal: In lawsuit over teen's death, judge rejects free speech rights for AI chatbots (23 May 2025)

About this article

This analysis was researched and written by The Corporate INTL Newsroom, which covers cross-border legal, regulatory and business developments for lawyers, professional advisers and financiers in over 150 jurisdictions. It has been checked against the complaint filed in the Northern District of California, the Government of British Columbia's statement, OpenAI's published correspondence and primary reporting. Every allegation described here is an allegation made in pleadings or public statements; none has been tested in court, and OpenAI disputes the claims made against it in related cases. This article is general information, not legal advice; for advice on a specific matter, consult a qualified adviser. Last reviewed 22 September 2026. For more analysis like this, visit the Corporate INTL newsroom or subscribe to Corporate INTL.