A 29-year-old Sydney man has been charged after allegedly telling an artificial intelligence chatbot that he intended to carry out a shooting at a childcare centre, according to reports of proceedings at Sutherland Local Court.
David Lan Thackray was arrested on Sunday morning after Interpol reportedly flagged his exchanges with Claude, an AI assistant developed by Anthropic.
He faces one count of using a carriage service to menace, harass or offend. The case raises questions about how conversations with AI services may become relevant to criminal investigations, even when users believe they are communicating privately.
At a court appearance on Monday, prosecutors opposed his release, warning of the potential consequences if he breached bail conditions. His lawyer, Danilo Rajkovic, argued that the remarks were made during a period of stress and frustration, and disputed that they reflected a genuine intention to harm anyone.
The judge ordered a mental health assessment before deciding whether to grant bail, with Thackray expected to return to court at a later date. The proceedings have not established that an attack took place or that the alleged plan was acted upon.
Sydney Court Hears Details of Alleged Claude Chatbot Threats
According to a police facts sheet tendered to Sutherland Local Court, Thackray allegedly told Claude he would 'go down to the local Westfield right now and shoot up the local daycare'. The court also heard that he referenced Naveed Akram, who has been charged in connection with the Bondi Beach attack.
During the bail hearing, prosecutors argued that the alleged remarks should be taken seriously, warning that there would be a 'catastrophic' outcome if Thackray were released and then breached his bail conditions.
'What he intends to do is start murdering people in Sydney just like the alleged terrorist did in Bondi,' prosecutors told the court.
They also said the chatbot exchanges offered insight into the concerns surrounding the alleged threats. 'We're understanding what's in his mind through his interactions with the chatbot,' they said.
These statements represent the prosecution's position, rather than a finding that Thackray intended to carry out the alleged attack. His lawyer challenged that interpretation, telling the court 'there was no weight' behind the comments.
Rajkovic said his client had turned to Claude while seeking help with an ear infection and subsequently went on what the lawyer described as a stress-induced tirade. He also argued that Thackray had not instructed the chatbot to send threats to any person or organisation.
According to the reports, Thackray has lived in Australia for more than 17 years, is studying international relations and has no criminal history, as stated by his lawyer. His Claude account is understood to have been suspended.
Anthropic's Claude Safety Rules and Police Data Requests
The case puts the handling of threatening conversations into focus, but it does not establish precisely which internal systems identified the alleged remarks or how the information was passed to Interpol.
Anthropic says its safeguards include monitoring for misuse, enforcing its usage policies and taking action against accounts that violate its rules. Its published transparency material says such action can include warnings, suspensions or termination of access.
Anthropic's published safety material also describes classifiers used to detect certain child-safety risks, although the available reporting does not establish whether any such system played a role in this case. Anthropic's transparency information outlines these measures.
The company also publishes a policy explaining when it may disclose user information to government authorities.
It says customer information is not disclosed in response to government requests unless there is valid legal process, such as a warrant or subpoena, with an exception where the company believes an emergency could result in imminent physical harm or death and prompt disclosure could help avert it.
Anthropic's government-request policy sets out those conditions.
That policy does not mean every troubling conversation is automatically reported to police. The circumstances of this case, including the precise process through which Interpol became involved, have not been fully established in the material available publicly.
For users, the distinction is important: an AI chat may feel like a private exchange, but the service's policies, applicable law and the nature of a potential emergency can affect how information is handled. Conversations with commercial AI services should not automatically be treated as legally privileged or guaranteed to remain confidential.
Thackray is expected to return to court following the assessment. The allegations and the competing accounts presented by the prosecution and defence remain central to the case.
