US used AI chats as evidence in criminal cases

Cases in the US, Canada and Australia show that conversations with AI chatbots can be used as evidence in criminal investigations. Experts warn that automated reports may lead to false alarms and privacy violations.

US used AI chats as evidence in criminal investigations

Investigation and court records in the US show that statements entered into AI chatbots can be used in criminal cases, while the privacy of these conversations is also being debated. Conversations on systems such as ChatGPT do not have the same absolute protection as attorney-client or doctor-patient communications.

Chat histories used in US cases

At Missouri State University, a 19-year-old student wrote to ChatGPT in August 2025 that he had damaged 17 vehicles in the campus parking lot at night and asked whether he would go to prison. Investigators accessed the exchange after the student consented to an examination of his mobile phone.

ChatGPT records were considered evidence against the student in court and influenced the sentence imposed. The case raised questions about the standards that should determine how statements in AI chats are handled in investigations.

Court records concerning two university students who went missing in the US and were later found dead also included a question from one student's roommate to ChatGPT about what would happen if a person were placed in a black garbage bag and thrown into a dumpster. ChatGPT responded that the idea appeared dangerous.

OpenAI sued over attack case in Canada

The government of British Columbia, Canada, sued OpenAI over a shooting in the province that killed nine people and injured 27. The government cited the failure to notify security forces about the attacker's conversations with ChatGPT.

ChatGPT arayüzü (arşiv)

OpenAI Chief Executive Officer Sam Altman apologized because police had not previously been informed about the attacker's conversations. The incident broadened the debate over whether existing standards are sufficient to determine when AI companies should inform security forces.

Research conducted in Australia also found that children who experience sexual abuse on online platforms are more likely to tell AI chatbots about it than to report it to authorized people. The findings brought to the fore the question of how chatbots should respond to sensitive disclosures.

Automated reports may generate false alarms

There is no common approach to whether AI chatbots should pass suspected crime plans or confessions to security forces without human review. Because people may express themselves in different ways, a conversation that algorithms classify as indicating criminal intent or a confession may appear harmless during human review.

It also emerged that Claude, an AI model developed by US-based Anthropic, made a false murder report to a US police hotline in July. The development increased questions about the reliability of chatbots' automated reports.

If systems identify potential criminals based solely on chat histories, they may overlook biological, psychological and socioeconomic factors that could affect a person's likelihood of committing a crime. If companies report conversations they consider suspicious too frequently in an effort to avoid liability, they may violate user privacy and increase the workload of law enforcement agencies.

Experts take cautious approach to crime prediction

Matthew Trail, a lawyer and research associate at Goethe University in Germany, said AI companies apply different safety rules. Trail said chatbots are trained not to respond to queries such as "planning a murder," while some topics, including child sexual abuse, are automatically referred to human moderators.

San Francisco'da OpenAI'ın genel merkezinin bulunduğu bina (arşiv)

Trail said users may still receive answers from chatbots if they present a harmful request as preparation for a work of fiction. He noted that AI companies may be asked for a person's chat history if law enforcement suspects that individual.

Arguing that the primary task of AI chatbots should not yet be to predict crimes that have not been committed, Trail said the field is new and that laws and policies have not yet fully adapted to the issue. Trail also said he was cautious about designing models to collect more data than they currently gather and make decisions of this kind.

New rules on conversation privacy under consideration

The growing use of AI chatbots to discuss personal matters such as health problems, psychological support and life advice has intensified the ethical debate over how these conversations should be protected. Sam Altman said during a program he attended in July 2025 that conversations with AI chatbots should be subject to a privacy principle similar to the confidentiality between a therapist and a client.

Robert Diab, a professor at Thompson Rivers University’s Faculty of Law, said there was no clear reason for laws that do not extend confidentiality to confessions between accomplices to recognize conversations between people and AI chatbots as communications requiring confidentiality.

Diab said leading models such as ChatGPT appear capable of detecting whether users are inclined toward crime and that companies could prevent some crimes by notifying authorities. However, he warned that a system requiring all suspicious conversations to be reported could increase the workload of law enforcement agencies because of false alarms.

Diab also said that even if security measures were developed, people who wanted to use AI to commit crimes could find a way around them.

Source: Anadolu Ajansı

Have a story, event or announcement of your own?Send us your news on WhatsApp

Spotted an error in this story?Report it on WhatsApp

Share on all social platforms