It's not clear from the letter when the new protocol took effect. But the company didn't flag Jesse Van Rootselaar to police when it banned her first account.

Canadian officials have been scrambling to figure out what OpenAI knew about the Tumbler Ridge, B.C., mass shooter, after a bombshell report released by the Wall Street Journal. It revealed that OpenAI had banned the shooter's ChatGPT account months before the tragedy, but decided not to report the account to authorities at the time.
OpenAI has said Van Rootselaar's activities didn't meet the company's threshold for informing law enforcement because it didn't identify credible or imminent planning at the time.

"With the benefit of our continued learnings, under our enhanced law enforcement referral protocol, we would refer the account banned in June 2025 to law enforcement if it were discovered today," O'Leary's letter reads.

Earlier this month, Van Rootselaar killed her mother and half-brother at the family home before going to the local secondary school, where she killed five students, an educational assistant and then herself.