OpenAI Banned Suspect’s ChatGPT Account Months Before Deadly Canada School Shooting
by Suraj Malik - 5 months ago - 5 min read
OpenAI banned a ChatGPT account belonging to the person responsible for the deadly Tumbler Ridge school shooting months before the attack, after its systems detected activity connected with violent scenarios.
The disclosure has intensified scrutiny over how AI companies identify serious threats, when they should contact law enforcement and whether banning an account is sufficient when employees believe online behaviour may indicate a risk of real-world violence.
Account was flagged in June 2025
OpenAI said it identified the account of Jesse Van Rootselaar in June 2025 through its abuse-detection systems. The company determined that ChatGPT had been misused in connection with violent activity and banned the account for violating its policies.
Employees reportedly considered referring the case to the Royal Canadian Mounted Police. OpenAI ultimately decided that the activity did not meet its threshold for reporting because it did not indicate an imminent and credible threat of serious physical harm.
Around eight months later, on February 10, 2026, the 18-year-old killed eight people in Tumbler Ridge, British Columbia, before dying by suicide. The victims included family members, students and a school employee.
OpenAI provided information to Canadian authorities after the attack and said it was cooperating with the investigation.
Canadian officials questioned OpenAI’s response
The case prompted Canadian officials to summon OpenAI safety representatives for discussions about the company’s escalation procedures.
Canada’s AI minister, Evan Solomon, said credible signs of serious violence should be escalated responsibly and argued that an internal company review was not enough when public safety could be involved. Officials said they were disappointed that OpenAI did not present more specific new safeguards during the meeting.
OpenAI’s decision illustrates the difficult balance platforms face between protecting user privacy and intervening when conversations appear dangerous.
Reporting every violent or disturbing conversation could expose users to unnecessary investigations, including people discussing fiction, journalism, personal fears or academic subjects. Waiting for an explicit plan, location or immediate timeline, however, may mean that companies act only after the clearest opportunity for prevention has passed.
A ban does not resolve the underlying risk
The most important issue is not whether OpenAI’s moderation system worked. It apparently detected the concerning activity and removed the account.
The harder question is what should happen after detection.
An account ban limits access to one service, but it does not necessarily reduce the user’s underlying intent or prevent them from moving to another chatbot, social platform or online community.
OpenAI’s system appears to have identified a safety concern, but the case shows that detection and prevention are not the same thing.
AI companies may now face pressure to create more detailed escalation systems between two extremes: taking no action beyond a ban and immediately reporting a user to police.
Such a system could consider repeated violent queries, movement from fictional discussion to operational planning, references to real people or locations, access to weapons and evidence that the threat is becoming more specific.
Human review would remain essential because automated systems cannot reliably determine intent from isolated text alone.
The case could reshape AI safety policies
The Tumbler Ridge attack is likely to influence future discussions about the responsibilities of generative AI providers.
Until now, much of the AI-safety debate has focused on whether chatbots refuse requests for weapons instructions, cyberattacks or other harmful activities. This case raises a separate issue: what platforms should do when repeated conversations create concern even when the model refuses to assist.
The distinction matters.
A chatbot can technically comply with its safety policy while the surrounding conversation still contains warning signs. That means safety cannot be measured only by whether the model produced prohibited instructions.
Companies may need clearer rules covering internal review, account restrictions, evidence preservation, senior escalation and referrals to authorities. Governments may also seek greater transparency about how frequently AI platforms detect serious threats and how often those cases are reported.
Our analysis
OpenAI’s decision should not be simplified into a claim that one police report would certainly have prevented the attack. Authorities were still investigating the motive and the suspect’s wider online activity, and it is impossible to know what would have happened under a different escalation decision.
However, the case exposes a clear weakness in current AI governance: companies have developed systems capable of detecting concerning behaviour, but the rules for acting on those signals remain largely private.
That creates three immediate questions for the AI industry:
| Safety question | Issue for AI providers |
|---|---|
| When does concerning activity become reportable? | Companies need criteria more detailed than a broad “imminent threat” standard |
| Who reviews high-risk cases? | Automated flags should be assessed by trained human safety teams |
| How transparent should platforms be? | Regulators may demand anonymised reporting on serious-threat escalations |
The wider lesson is that content moderation cannot end at removing access.
As AI platforms become places where users disclose intentions, fears and fantasies, providers will increasingly be expected to distinguish between disturbing speech, policy violations and credible indications of real-world danger.
For OpenAI and its competitors, that may become one of the most difficult safety responsibilities attached to operating a widely used conversational system.