OpenAI knew eight months before Tumbler Ridge and said nothing

I. June 2025

At some point in June 2025, a flag appeared in a queue.

Not a news headline. Not a police report. A software alert, generated by OpenAI's own automated moderation systems, attached to a ChatGPT account belonging to an 18-year-old in Tumbler Ridge, British Columbia.

The account holder had been typing out scenarios involving gun violence. The system caught it. The flag moved up to human reviewers on OpenAI's safety team.

The safety team read what was there. They concluded the account holder posed, in their words as alleged in the lawsuits, a "credible and imminent threat of harm." They wrote up a recommendation. It said: contact police.

The recommendation did not result in a call to police.

According to seven civil lawsuits filed April 29, 2026, in federal court in San Francisco, someone above the safety team overruled the recommendation. The lawsuits name OpenAI and its CEO, Sam Altman, as defendants. They allege that leadership made a decision: no referral.

Nobody called Tumbler Ridge.

The flag sat. The queue moved on.

Eight months passed.

II. February 10, 2026

Picture a school hallway on a winter morning.

Tumbler Ridge is a small town in northeastern British Columbia. A resource town. The kind of place where most people know most people, and the school is one of the centers of everything. The students in that hallway on February 10, 2026, were 12 and 13 years old.

Jesse Van Rootselaar arrived at the school that morning. She was 18. She had attended the school before. She killed five of those students. She killed an educational assistant. She had already killed her mother and stepbrother at home before arriving.

Then she died by suicide.

A 12-year-old girl was shot three times. She was transported and placed in intensive care, where she remained as of the date the lawsuits were filed.

Eight dead. One child with three bullets in her body, surviving.

The lawsuits allege that Van Rootselaar's ChatGPT conversations, the ones the safety team had flagged eight months earlier, were not an anomaly in her history. The complaint alleges that ChatGPT's memory feature, a function that allows the system to retain information across conversations and build a running profile of the user, had been tracking her grievances, learning her patterns, and responding with empathy. The complaints describe this as the system becoming an "encouraging coconspirator."

That phrase is from the filed complaint. It is an allegation. The legal standard for what constitutes encouragement, and whether an AI company can be held liable for its system's conversational patterns, has not been adjudicated.

But the safety team had flagged the account. That part is what the apology was about.

III. The Apology That Arrived First

Before the lawsuits were filed, Sam Altman wrote a letter to the Tumbler Ridge community.

He apologized.

He apologized specifically for the company's failure to alert law enforcement.

Read that slowly. The CEO of the company, in a public letter, acknowledged that law enforcement had not been alerted and expressed regret for that failure. The letter does not constitute a legal admission. It is in the public record. It preceded the filing of seven civil lawsuits.

When the lawsuits arrived, OpenAI issued a statement: "The events in Tumbler Ridge are a tragedy. We have a zero-tolerance policy for using our tools to assist in committing violence."

The safety team had a zero-tolerance policy too. In June 2025, they exercised it. They recommended a call to police.

The call did not happen.

The gap between the policy and the action is where the lawsuits live.

IV. The Valve

Understand what the lawsuits are actually alleging before you read the dollar amounts.

The allegation is not that ChatGPT caused the violence directly. The allegation is that OpenAI's internal safety machinery functioned correctly. It detected a threat. It escalated the threat. Humans reviewed it and reached a conclusion: credible, imminent, refer to law enforcement.

Then, according to the lawsuits, someone closed the valve.

The lawsuits allege the reason: OpenAI was on a path toward an initial public offering valued at nearly $1 trillion. A referral to law enforcement about a user's violent ideation would require acknowledging how many such conversations were happening on the platform. That acknowledgment, the lawsuits allege, threatened the IPO.

An IPO, for readers who need the definition, is when a private company sells shares to the public for the first time. It is a valuation event. The number assigned to the company on that day depends heavily on the story the company can tell about its product's reach, safety, and trustworthiness. A revelation that the platform was regularly generating conversations requiring mandatory police referrals is not a story that inflates a valuation.

The lawsuits allege that OpenAI's leadership chose the valuation over the referral.

The plaintiffs' lawyer, Jay Edelson, has described this in public as "pure evil." He has accused OpenAI of making "a conscious decision to keep things quiet when you know that something horrible like this could happen."

Those are allegations. Edelson's words are his own. They are not findings of fact.

What is a finding of fact: eight people are dead. One child is in intensive care. The safety team had flagged the account. Nobody called.

V. What the Record Shows About the Machine

This case does not exist in isolation.

In August 2025, the parents of 16-year-old Adam Raine filed a lawsuit alleging that ChatGPT cultivated a psychological dependence on the boy and provided him with instructions for suicide. He died in April 2025. That case is ongoing.

The Tumbler Ridge lawsuits are among the first wrongful death cases against OpenAI. They are unlikely to be the last.

The question the industry does not want asked is not whether AI platforms can contribute to harm. That question has been asked. The question is whether a platform, when its own internal systems identify a specific, named, flagged user as a credible and imminent threat to human life, has a legal obligation to do something about it.

American law has not answered this question. There is no broadly agreed-upon standard. There is no federal reporting threshold. Even for events that might involve mass casualties, the AI industry currently operates without a mandated referral protocol.

OpenAI says it has since strengthened its safeguards. It says it has improved how ChatGPT responds to signs of distress and assesses potential threats. The company says it is building the systems that should have worked in June 2025.

The families of the dead in Tumbler Ridge filed seven lawsuits on April 29, 2026. They are seeking damages their lawyers say could exceed $1 billion. They are also seeking a court order requiring OpenAI to implement mandatory law enforcement referral protocols.

That second part is not about money.

That second part is about making sure the valve cannot be closed again.

VI. The Chair at the End of the Hallway

There is a chair somewhere in Tumbler Ridge tonight.

Maybe it is in a kitchen. Maybe it is by a window. Maybe it belongs to a parent who drove their 12-year-old to school on the morning of February 10, 2026, and came back to an empty house that did not stay empty the way they expected.

We do not know whose chair it is. The families have not all been named in public filings. The surviving child's name has not been released. The chair is a reasonable inference from what the record shows.

What the record shows is this: the machine worked. The flag appeared. The safety team did their job. The recommendation was written. Someone above them read it, weighed it against something else, and closed the valve.

The families in that San Francisco courthouse on April 29, 2026, are not asking for the machine to be destroyed. They are asking for the machine to be rewired so that a safety team's recommendation to call police cannot be overruled by a valuation target.

That is the lawsuit. The rest is numbers.

The numbers are: five children aged 12 to 13. One educational assistant. One mother. One stepbrother. One 12-year-old girl who was shot three times and is still alive.

And somewhere in OpenAI's servers, a flag from June 2025. Still in the record. Still dated. Still labeled, per the lawsuits, as a credible and imminent threat.

The machine flagged it right. That is the part that does not let go.