That detail is both serious and, for now, disputed. TechCrunch reported that the complaints provide no direct evidence that Lehane personally overruled the safety team. Plaintiffs’ lawyer Jay Edelson said the claim rests partly on OpenAI’s organizational structure and his firm’s wider investigation. Discovery would be expected to test whether internal messages and decision records support it. Corporate charts can suggest who had authority. They are less effective at proving who actually said no.
The complaints also allege OpenAI merely deactivated Van Rootselaar’s account instead of imposing a system-wide ban. That allegedly allowed the attacker to return using another email address. Plaintiffs argue that continued access to ChatGPT amounted to substantial assistance and helped enable the planning and execution of the attack.
How has OpenAI answered the allegations?
OpenAI chief strategy officer Jason Kwon called the lawsuits’ account of the company’s safety process false. In a post on X, he said the allegation that Lehane intervened was “absolutely false.”
Kwon also rejected claims that political or public-relations concerns influenced employees handling dangerous-user reports. He said it was “completely untrue” that the people making those decisions failed to prioritize safety.
At the same time, Kwon acknowledged the difficulty of deciding when threatening chatbot conversations justify contacting law enforcement. He framed the issue as a balance between identifying genuine threats and protecting the privacy of more than one billion users. He also said there is not a day when he does not think about Tumbler Ridge.
British Columbia Premier David Eby has been considerably less forgiving. After Altman apologized, Eby described the statement as “necessary, and yet grossly insufficient.”
Edelson said the new plaintiffs include children who watched teachers die and believed they would be killed. Their cases shift the litigation beyond relatives of those who died, adding claims from people who survived or witnessed the violence inside the school.
Why is the legal standard difficult?
The plaintiffs are pursuing an aiding-and-abetting theory, which generally requires more than showing that a company acted carelessly. They must establish that OpenAI intentionally provided meaningful assistance connected to the attack.
That makes several factual questions central to the cases:
- What exactly did Van Rootselaar tell ChatGPT?
- How did OpenAI’s review systems classify those conversations?
- Which employees recommended notifying Canadian authorities?
- Who had the power to approve or reject that recommendation?
- Did access to ChatGPT materially assist the attack?
The complaints offer an account of internal events, but some of the most consequential claims appear to rely on secondhand information. OpenAI’s records, employee testimony and internal chain of command could therefore determine whether the cases survive early dismissal efforts.
The company also faces a lawsuit from the state of Florida accusing it of aiding and abetting mass shooters, including the attacker responsible for the 2025 shooting at Florida State University. That case adds pressure to a broader legal question: when an artificial intelligence system detects apparently credible threats, what duty does its operator have to intervene?
Will the cases remain in California?
OpenAI has asked a California federal judge to dismiss the seven earlier Tumbler Ridge cases so the litigation can proceed in Canada, according to MLex. No ruling has been reported.
The company argues that the victims, police officers, medical providers, school officials and much of the evidence concerning causation are in British Columbia. Many of those witnesses and records sit beyond the California court’s subpoena power.
The plaintiffs, however, sued in the state where OpenAI is based and where decisions about its safety procedures may have been made. That creates a jurisdictional split between the location of the alleged corporate conduct and the location of the physical harm. Corporate location and evidentiary location, inconveniently, are not the same question.
The venue fight could shape how easily each side obtains documents and testimony. It does not answer whether OpenAI assisted the attacker, but it may decide which court gets to examine that claim and under which country’s legal framework.
For now, the central allegations remain unproven and sharply contested. The next phase will test whether the plaintiffs can obtain evidence supporting their account of OpenAI’s internal decisions, rather than relying mainly on what the company’s structure suggests may have happened.
