Man plotted daycare shooting in Claude chat, arrested in Sydney

What's going on

In Sydney, Australia, a 29-year-old man was arrested after telling Anthropic's AI assistant Claude he wanted to kill children at a daycare center near his home. That's according to the Sydney Morning Herald, which Futurism, The Independent and the broadcaster SBS cite. English-language reports name the man as David Lan Thackray (SBS spells it "Sacray"). He was arrested on Sunday, October 18.

What happened

According to court documents, he wrote to Claude that he wanted to "go to the local Westfield right now and shoot up the local daycare." Westfield is a shopping center. The daycare is in the Hurstville area of southern Sydney, where he lives. He also allegedly mentioned Naveed Akram in the conversations. Akram is charged over his alleged involvement in the December 2025 attack on a Hanukkah celebration at Bondi Beach, in which at least 15 people died. According to the Independent, his father, Sajid Akram, was shot dead by police.

Prosecutors told the court: "What he intends to do is murder people in Sydney, just like the [alleged] terrorist in Bondi." The chats, they said, show what is going on in his head. Releasing him on bail could have "catastrophic" consequences. The Independent reports that his Claude account was subsequently banned.

Charge and defense

He is charged under an Australian law with using a "carriage service" (a communications service) to threaten, harass or offend. His lawyer, Danilo Rajkovic, argues that his client was just venting. He was stressed and frustrated because he had been looking for help with an ear infection. His words carry "no weight", the lawyer says. He did not instruct Claude to threaten anyone and believed the conversation was private. Thackray studies international relations, has no criminal record and has lived in Australia for more than 17 years. The court in Sutherland remanded him in custody for a psychiatric assessment. The case will return to court afterward.

Open questions: how did Interpol get the chats?

The reports say the trigger was that Interpol flagged the conversation and notified Australian police. Futurism stresses that it is unclear how Interpol gained access and whether Anthropic passed the chats on. Anthropic says generally that its models are trained to refuse requests to plan or incite violence, and that automated safety systems are meant to detect and block dangerous conversations. Where necessary to prevent emergencies involving danger to life or serious bodily harm, user data may be shared with law enforcement. The Independent also points to an Anthropic report from September: between December 2025 and August 2026, its threat intelligence team detected and disrupted abuse operations in seven areas. These include cyber operations, influence operations, surveillance, fraud, biological misuse, development of conventional weapons and distillation (siphoning off a model's knowledge to train other models).

Background: chatbots and violence

Futurism notes that chatbots have become confidants for people in distress, including those with delusions or fixations. Heavy chatbot use has been linked to crimes, including a murder-suicide and several mass shootings, such as the 2025 shooting at Florida State University and one at a school in Tumbler Ridge, British Columbia, in February. There is no industry standard for when, or whether, AI companies should notify authorities. Chats are often deeply intimate, which raises fears of surveillance and privacy violations. At the same time, they could provide clues that help prevent violence.

Our take

The case shows that conversations with AI assistants don't necessarily stay private when they contain signs of concrete violence. Whether the statements reflected real intent or mere frustration is an open question, and the court and the assessment will settle it. For now, the presumption of innocence applies. The path the data took from Anthropic to the police is also undocumented. The trade-off between prevention and privacy remains a critical issue, as does the risk of false alarms. For users, the lesson is this: if you write threats or violent fantasies into a chatbot, expect to be banned and reported.