▲ Claude logo
A man in his 20s in Sydney, Australia, has been arrested by police after discussing a plan to carry out a mass shooting at a childcare center with the AI chatbot Claude.
According to local media outlets including The Sydney Morning Herald, Sydney police arrested 29-year-old David Lan Sacray on the morning of the 18th (local time).
He is accused of revealing to Claude a plan to shoot children at a childcare facility near Hurstville in southern Sydney, where he lives.
According to police documents submitted to the court, he told Claude, "I am going to go straight to the local Westfield (shopping center) and commit a mass shooting at the childcare facility there."
It is also reported that during his conversation with Claude, Sacray mentioned Naveed Akram, one of the perpetrators of the Sydney Hanukkah shooting last December that left 15 victims dead.
He was caught after Interpol intercepted the conversation and alerted Australian police. Prosecutors have charged him with using a telecommunication service to menace, harass, or cause offense.
"We have an insight into what is inside his head through his conversations with the chatbot," prosecutors emphasized, adding that he intended to "kill people in Sydney."
They added that releasing him on bail and any subsequent violation of bail conditions could lead to "catastrophic" results, and therefore he should remain in custody.
In response, Sacray's defense lawyer argued that Sacray was merely venting out of stress and frustration while talking to Claude to treat an ear infection, and that his words carried no real intent or weight.
The lawyer also added that Sacray, an international relations student, has no criminal record and has lived in Australia for over 17 years.
The local court remanded him in custody the previous day for a mental health assessment, and plans to call him back to court for a hearing once the assessment results are available.
According to Anthropic, the developer of Claude, its AI models are trained to refuse requests to plan or incite violence, and feature automated safety systems designed to detect and block dangerous conversations.
In particular, when deemed necessary to prevent emergency situations that could endanger human life or cause serious physical injury, user data may be shared with law enforcement agencies.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.