Artificial intelligence chatbots are becoming unwitting witnesses to crimes around the world, yet the legal frameworks needed to govern their role in criminal proceedings remain far behind technological developments. Chat logs are now being admitted as evidence in courts, but questions over the chatbots' own legal obligations persist.
At Missouri State University in the United States, a 19-year-old student confessed to ChatGPT in August 2025 that he had damaged 17 cars in a campus parking lot overnight, asking the chatbot whether he would go to jail. The chat history, retrieved after the student consented to a phone search, was used as evidence against him in court. In a separate US case, a roommate of two university students who were found dead asked ChatGPT what would happen if a person were put in a black garbage bag and thrown in a dumpster — days before the bodies were discovered. The AI responded only by noting that the idea "sounded dangerous."
The government of British Columbia, Canada, filed a lawsuit against OpenAI in February, alleging the company failed to alert authorities to the attacker's ChatGPT conversations before a mass shooting that left nine dead and 27 injured. OpenAI CEO Sam Altman publicly apologised for not having informed police sooner. Meanwhile, research conducted in Australia found that children who experienced online sexual abuse were more likely to disclose it to AI chatbots than to authorities.
Unlike communications between lawyers and clients or doctors and patients, conversations with AI chatbots carry no legal confidentiality protections, and companies can share chat histories with law enforcement. The risks of automated crime reporting were highlighted when Anthropic's Claude model was found to have made a false murder report to a US police line in July.
Matthew Trail, a US legal researcher at Goethe University in Germany, noted that AI companies follow varying internal rules and that chatbots are trained to avoid answering queries such as requests for help planning a murder. Trail cautioned that predicting future crimes should not be a core function of these systems, saying, "This is a very new situation. Laws and policies have not yet fully adapted to this issue." He also warned that chatbots relying solely on conversation histories cannot account for the biological, psychological and socioeconomic factors that contribute to criminal behaviour.
Professor Robert Diab of Thompson Rivers University Faculty of Law in Canada argued that existing laws, which do not extend confidentiality to conversations between co-conspirators, provide no reason to treat AI chatbot conversations as confidential. Diab suggested that leading models appear capable of identifying users who may be inclined toward criminal behaviour, but warned that mandatory reporting requirements could lead to over-reporting, violating user privacy and burdening law enforcement. OpenAI's Altman, speaking in July 2025, called for chatbot conversations to be subject to confidentiality principles similar to those governing therapist-client relationships.
