Product1 publisher3 min readPublished
British Columbia sues OpenAI over the police call its reviewers recommended
The province's negligence claim turns on an internal recommendation to notify the RCMP about the Tumbler Ridge shooter's chats. For anyone shipping a conversational product, the review queue is now a discoverable record.
The Product Desk · Product desk
What happened
- British Columbia has sued OpenAI over the company's failure to notify authorities about the conversations the Tumbler Ridge shooter had with its chatbot.
- The suit says the shooting could have been prevented had OpenAI heeded its own human reviewers' advice to notify the Royal Canadian Mounted Police about the shooter's activity on the service.
- The province also wants OpenAI to pay for the construction of a new school in Tumbler Ridge.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure Whoever runs a trust-and-safety queue now owns discoverable evidence of knowledge: the reviewer's note, its date, and whether the response stayed inside the product.
- decision Any team shipping a conversational product has to settle in writing what threshold sends a chat to police and who is authorised to make that call at night, because OpenAI's own pledge to report credible threats is the yardstick being applied.
- precedent A province suing in the vendor's home court over an escalation failure gives other governments a venue to copy without passing new law first.
- contradiction The product fix and the relationship are being judged separately: OpenAI's commitment to report credible threats does not answer the province's allegation that no meaningful safety engagement followed the shooting.
Banning is the action review tooling makes easy. One click, ticket closed, reviewer moves to the next item. Phoning a police force in another country is not a button in the queue, and it was not what happened after OpenAI employees flagged the account in 2025 for potential warnings of committing real-world violence [4]. The account was banned. The shooter later opened a second one, which OpenAI discovered afterwards [5].
Teams describe that queue internally as a moderation function, measured in tickets cleared and accounts actioned. British Columbia is describing the same queue as a record of what the company knew.
"In any other circumstance where a person aids, encourages or conspires to commit a criminal offense or is criminally negligent, they can be investigated, prosecuted and judged," said Niki Sharma, the attorney general of British Columbia [9]. "No company should be exempt from accountability simply because the technology is new," she said [10].
A failure-to-notify claim is easier to document than a claim about what a model said. An output claim needs an argument connecting generated text to an act. The escalation claim needs an internal note with an author and a date, and the province says its reviewers' advice to notify the Royal Canadian Mounted Police was given and not followed [3]. According to Engadget, the complaint pleads both theories: unsafe design, and negligence for not alerting law enforcement [2]. The complaint text is not in the public record here, and Engadget's account of it rests on reporting by the Wall Street Journal and the New York Times [2].
The venue is a deliberate choice. Sharma's office filed in California, OpenAI's home, instead of Canada, which Engadget reports could mean larger payouts for the victims' families [12]. The province also wants OpenAI to pay for the construction of a new school in Tumbler Ridge [13]. That remedy is priced by a building, not by a per-user damages model.
OpenAI has committed to notifying authorities about credible threats, and Sam Altman apologised for not alerting them when the first account was banned [6]. Premier David Eby called the apology "necessary, and yet grossly insufficient" [7]. The suit also alleges the company has not meaningfully engaged with Canada and British Columbia on safety since the shooting [8].
Two counts from a team's own logs place it against this complaint. First, how many conversations a human reviewer flagged in the past year as containing a potential warning of real-world violence. Second, how many of those produced a communication to anyone outside the company: a police force, a school district, a regulator. When the first number is not zero and the second is, the record matches what the province alleges about OpenAI [3]. The person who has to answer for that gap is whoever owns the escalation policy, and the policy has to name a threshold, a recipient, and someone on call at 2am who is allowed to make the call.
What to watch
- Whether the California court keeps the case or sends it back to Canada on forum grounds.
- Whether the filed complaint, once public, dates the reviewers' recommendation and names who at OpenAI saw it.
- Whether OpenAI publishes the threshold and recipient list behind its commitment to report credible threats.