If a conversation with an AI chatbot suggests that someone is planning a killing or a major attack, why are the police not told? Should AI companies put the privacy of personal conversations first, or the prevention of a potentially deadly attack? The question has been raised again after it emerged that the attacker in a school shooting in Canada had used ChatGPT beforehand. Technology companies, lawyers and researchers are divided over whether law enforcement should be informed when violence is planned in conversation with an AI.
On 10 February 2026, Jesse Van Rootselaar, 18, carried out a deadly shooting in Tumbler Ridge, British Columbia, Canada. The attacker first killed their own mother and stepbrother at home, then went to a school in Tumbler Ridge and killed six more people, including five students, before taking their own life. It was the deadliest school shooting in Canada in nearly four decades.
According to reports, Van Rootselaar had held conversations with ChatGPT about mass killing before the attack. On 14 October, US authorities charged a man from Washington state with conspiring with Van Rootselaar to plan the attack.
The case has raised questions about the role of AI companies. In September, the British Columbia government sued OpenAI and its chief executive, Sam Altman, alleging that the company did not take adequate steps to prevent the attack and that it knew about Van Rootselaar’s conversations. Family members of those killed have filed separate lawsuits. OpenAI’s leadership has publicly expressed regret for failing to alert the authorities.
OpenAI did not respond to a request for comment from Al Jazeera.
Opinion is divided, however, on whether the police should be told whenever a conversation hints at violence. Civil rights lawyer Jenin Younes said mandatory surveillance and disclosure rules could raise serious concerns for civil liberties and constitutional rights. If people fear that their private or uncomfortable thoughts could be reported to the police, many may avoid using AI chatbots.
On the other hand, AI can provide information that could help someone carry out a major deadly attack. People involved in the debate therefore say it is urgent to settle how far AI companies are responsible for potential attacks.
In August, the AI company Anthropic alerted San Francisco police after a user’s conversation suggested a threat to kill its chief executive, Dario Amodei. According to reports, the user mentioned an AR-15 rifle and claimed to have Amodei in his sights. The man later told the San Francisco Standard that he had been joking and had no intention of hurting anyone. He was not detained and no charges were brought against him.
Fazl Barez, an AI safety researcher at the University of Oxford, said current AI models cannot reliably tell the difference between a fictional discussion and a credible plan to commit real violence. If the threshold for alerts is set too low, the conversations of writers and researchers could be wrongly flagged as suspicious. Identifying a credible threat may also require reviewing the pattern and continuity of several conversations.
Experts say that a violent thought that comes to mind unbidden is not the same as planning an attack. Before treating a conversation as dangerous, it is essential to understand the person’s intent, the context of what was said and the pattern of their behaviour.
Chinmayi Sharma, a law professor at Fordham University, said the duty of AI companies bears some resemblance to the duty of mental health professionals to warn of potential danger. But AI companies do not have the same kind of trained staff, clear standards or enforceable rules. That creates a risk of bias in deciding which conversations are reported to the police, and the words of minority communities in particular could be viewed with excessive suspicion.
According to Bryan Choi, a law professor at the University of Colorado Boulder, a legal duty to warn of potential danger can apply in some situations in the United States. Under current legal precedent, however, it is unlikely to apply to OpenAI, because a chatbot is not a licensed psychiatrist and the company has no special relationship with potential victims. There may be scope for legal action over negligence or product liability, though that would be hard to prove.
As a possible solution, experts have proposed that an independent body set the standards for identifying threats and reporting them to the authorities. In their view, only the necessary information should be shared with law enforcement, so that irrelevant personal information stays private. Users should also be told clearly when their conversations may be monitored and in what circumstances that information may be disclosed to others.
Still, there is no generally accepted answer yet to the question of when AI companies should tell the police about signs of a plan to kill, and when they should protect a user’s privacy.
বাংলায় মূল প্রতিবেদন পড়ুন · Read the original Bengali report