British Columbia asks a California court to write OpenAI's escalation policy

The classifier fired. After the Wall Street Journal reported on the case, OpenAI said the account had been flagged by systems that identify "misuses of our models in furtherance of violent activities," and that the issues did not meet its internal criteria for reporting to law enforcement ^[11].

Two decisions sat downstream of that flag. Both were policy. The first is the threshold: OpenAI has said it notifies law enforcement when conversations suggest "an imminent and credible risk of harm to others" ^[13]. The province's complaint, citing OpenAI whistleblowers who spoke to the Journal, says safety team members recommended contacting police and that Sam Altman and other leadership overruled them ^[9].

The second is enforcement scope. The shooter's account was deactivated, and she obtained a new account and kept using the platform to plan the attack ^[10]. Deactivation is a credential-level control. Making it person-level means retaining identifiers from the first signup, matching them against later ones, and absorbing the lockouts that follow when the match is wrong. None of that appears in a model card.

The flag is dated June 2025 ^[9]. Police say the attack happened on February 10, when Jesse Van Rootselaar shot her mother and stepbrother at home before killing an educational assistant and five students aged 12 to 13 at her former school ^[8]. The earliest February that can follow a June 2025 flag is February 2026, so at least eight months separate the two events ^[19].

British Columbia is seeking damages for what it has spent and will spend on recovery, plus an order directing changes to the way the company handles ChatGPT conversations that could lead to violence ^[3]. An order like that has to be written in specifics: which classifier outputs escalate, to whom, on what clock, and whether a banned user's re-registration counts as continuing knowledge. The province filed in federal court in San Francisco, naming OpenAI and Altman as defendants ^[2].

Its damages claim reaches past wrongful death. The lawsuit counts the demolition of the school, the replacement building, and mental health care the province has absorbed ^[7]. More than 30 family members of victims and others affected have filed their own suits in California federal court, accusing the company of opting out of a chance to prevent the shooting ^[18]. Florida sued OpenAI in June, accusing it of misrepresenting ChatGPT's safety and of harming children by providing information to school shooters ^[12]. OpenAI has denied the claims ^[13].

Drew Pusateri, a spokesperson for OpenAI, called the Tumbler Ridge shooting an unspeakable tragedy ^[16]. "OpenAI remains committed to working collaboratively with government and law enforcement officials, and continuing to advance our ongoing safety work," Pusateri said ^[17]. Altman published a letter to the community saying he was "deeply sorry" OpenAI had not contacted law enforcement, and the lawsuit says he promised reforms and never followed through, despite efforts by British Columbia's attorney general to engage ^[6].

For this to become a build requirement elsewhere, a vendor has to be in the same position OpenAI was: conversations retained, a violence classifier scoring them, and a human queue behind the score. A duty to warn reaches only the vendors that kept something to warn from. The report does not say what OpenAI's internal reporting criteria require ^[15].