MAIN STORY
1. A tragedy exposes an unresolved duty
On February 10, 2026, attacks at a home and a secondary school in Tumbler Ridge killed eight victims; the attacker also died. Reuters and AP have reported that safety systems at OpenAI had previously flagged troubling interactions involving Jesse Van Rootselaar. The issue is not a proven claim that the chatbot independently caused the attack, but whether information already identified should have led to a timely warning.
In September, British Columbia filed a case in California alleging OpenAI failed to notify law enforcement about threats on its platform. Victims’ families filed separate lawsuits. These are allegations under legal scrutiny, not findings of liability. After the tragedy, OpenAI outlined measures including improved detection of repeat harmful activity, a direct contact channel with Canadian law enforcement and revised reporting thresholds.
2. Distinguishing dangerous intent from difficult words
Oxford University AI safety researcher Fazl Barez has cautioned that current models cannot reliably identify whether violent language is hypothetical or signals intended action. The same terminology might occur in a novelist’s draft, a legitimate classroom study, a distressed person’s thoughts or a credible threat. A phrase extracted from its setting can be dangerously misleading.














