An investigation by The Observer and Tortoise has reported that OpenAI's small team responsible for reviewing flagged ChatGPT conversations faced a backlog stretching back months, even as the chatbot grew to 1.2 billion weekly users. Based on interviews with current and former employees, lawyers and mental health professionals, the investigation raises questions about whether the company's safeguards kept pace with the intimate role ChatGPT now plays in many people's lives.
At the centre of the report is a former member of OpenAI's intelligence and investigations team, identified only as Sarah, who spoke on condition of anonymity because she feared retaliation. Reviewers on the team read conversations flagged by an automated system trained to spot dangerous behaviour, and could ban accounts or refer imminent threats to the authorities.
A backlog measured in months
According to the investigation, the team working through that caseload was small: between 12 and 60 people, based on sources who spoke to The Observer, although the newspaper said it could not confirm the exact size. Two current employees said the team remains small, and OpenAI itself has described the team as small.
Sarah described one case she could not forget: a teenager who had told ChatGPT he was being bullied and felt isolated, and whose conversations gradually turned to plans for violence against classmates. She spent a week reading several months of his chat logs, and only hours into the review realised he had named a date for an attack. By the time she reached the case, that date had already passed. She later found messages in which he expressed regret at not carrying out the plan, and as far as she is aware, he never did. The team did not operate a round-the-clock emergency rota, the report adds; instead, Sarah said, she had a phone she was expected to answer if an emergency arose during the night.
The Tumbler Ridge case
The report draws a parallel with Tumbler Ridge, a mining town in British Columbia, Canada, where an 18-year-old killed her mother and 11-year-old stepbrother at home in February 2026, then killed five students and a teaching assistant at the town's secondary school before taking her own life. Eight months earlier, when she was 17, OpenAI's automated systems had flagged her conversations with ChatGPT. According to The Wall Street Journal, around a dozen members of the intelligence and investigations team were aware of the concerning exchanges, and some believed police should be notified, but senior leadership decided the threat did not meet its threshold for alerting law enforcement, and the account was banned instead.
In April, OpenAI chief executive Sam Altman wrote to the people of Tumbler Ridge to say he was deeply sorry the company had not alerted police. The company told the Canadian government that under a revised protocol it would refer the shooter's first account to police for further investigation.
Lawsuits mount as OpenAI defends its safeguards
On 14 October, eight US states sued OpenAI, accusing the company of misleading the public about ChatGPT's safety. That case sits alongside at least 30 lawsuits brought by families and individuals alleging ChatGPT contributed to deaths and serious psychological harm, and a further 45 lawsuits from survivors and relatives of school-shooting victims in Florida and Canada alleging OpenAI failed to alert police after its systems identified disturbing conversations.
OpenAI said ChatGPT is trained to recognise mental distress, de-escalate crises and encourage real-world support. In a statement quoted in the investigation, a spokesperson said: "What happened in Tumbler Ridge was an unspeakable tragedy, and our thoughts remain with the victims, their families, and the entire community."
