OpenAI's review system flagged the shooter. Its leaders allegedly stayed silent

For OpenAI, the most dangerous evidence in these cases may be proof that its own systems worked. Thirty new lawsuits filed Wednesday in a California federal court accuse the company and its CEO, Sam Altman, of providing "substantial assistance and encouragement" to the suspect in the Tumbler Ridge school shooting in Canada, as The Verge reports. The plaintiffs were in the school during the attack: students, teachers, and the principal.

The complaints do not describe a moderation breakdown. They describe a moderation success followed by an alleged silence, and that inversion is what makes these cases hardest for OpenAI to answer in public. A company can blame a flawed model. It has a harder time explaining why no one acted after its own flag went up. The company has acted on internal warnings before: it paused work on Astra, its in-development model, after evaluations could not rule out the model hacking hardened systems unaided, per our report on that pause.

A flag that led to an alleged silence

OpenAI's automated review system, the lawsuits claim, flagged conversations about gun violence that the alleged shooter, Jesse Van Rootselaar, had with ChatGPT. The complaints say the safety team recommended contacting Canadian authorities. Instead, they allege, chief global affairs officer Chris Lehane took part in the decision to stay quiet, which they tie to concern about the company's "reputational and financial standing."

Deactivating an account, not banning a user

There is a second, more operational allegation in the filings, and it comes down to a single choice. OpenAI, the complaints say, deactivated an account instead of issuing a system-wide ban. Van Rootselaar could then sign up again with another email and keep using ChatGPT.

The complaints turn that detail into the legal core of the case. "Providing the Shooter with continued access to this tool constituted substantial assistance in itself," they allege, adding that the access "was a key factor in causing the Shooter to plan and carry out the attack."

The denial, and cases building elsewhere

OpenAI rejects the picture of its internal process. Chief strategy officer Jason Kwon said in a post on X, quoted by The Verge, that the filings contain "false claims" about the company's safety teams. It is "completely untrue," he wrote, "to say that the people at the center of these challenging decisions do not prioritize safety, or that there are 'political' or 'public relations' factors at play."

Kwon's answer defends the people who make these calls. The complaints go after the system that routes calls to them: what gets flagged, what gets escalated, and what happens when the two diverge.

The new cases are not a first attempt. Victims' families filed similar lawsuits in April, and the state of Florida is pursuing its own claim that OpenAI aided and abetted mass shooters, including the suspect in last year's attack at Florida State University. Every version rests on the same proposition: a company can be liable when its product and its moderation choices substantially help someone commit violence.

That proposition separates these cases from lawsuits about what a model says. OpenAI is already defending one of those over ChatGPT Health, where its own lead walked back the launch claims, as covered in our report. Nobody here argues the model should have refused a prompt. The argument is that OpenAI saw a high-risk user, had evidence of what he was planning, and allegedly chose its reputation over a call to Canadian authorities.

Read together, the complaints describe a moderation chain that failed at its last link. The model did not have to produce anything new. OpenAI, the plaintiffs argue, needed only to avoid putting the same tool back into the hands of a person its own review had already flagged.

What is missing from both the filings and the denial is the inside record: who saw the flagged conversations, when the safety team's recommendation reached the executives, and what was weighed against it. OpenAI says safety came first. Its published roadmap carries the same priority, while calling the safety part the harder one. The plaintiffs say it was traded away for business reasons. A California federal court now holds the argument.