February 26, 2026
soul – Police are increasingly examining suspects’ history of using generative artificial intelligence as key evidence to establish intent and motive, according to media reports citing legal experts on Wednesday.
In a recent case, after investigating the chat logs of OpenAI’s ChatGPT for a female suspect charged in a serial murder case in Gangbuk-gu, Seoul, investigative authorities announced that they had decided to pursue her with murder instead of manslaughter.
The suspect, a woman in her 20s named Kim, was charged with murder, aggravated assault, and drug violations. She is accused of giving drug-laced hangover pills to three men at a motel between December and February 9. Two victims died, but the first one survived with injuries.
According to police, the suspect asked ChatGPT, “Will people die if they take sleeping pills with alcohol?” Investigators considered this evidence of criminal intent.
Legal experts say such searches are becoming more common. Lawyers noted that authorities are increasingly examining AI chat logs generated during forensic analysis of mobile phones.
One lawyer, speaking on condition of anonymity, said the change is impacting defense strategy. “Now, when I work on a case, I review ChatGPT conversations with clients,” he said.
Experts point out fundamental differences between traditional browser searches and AI-powered conversations. While both can be used to seek information, AI conversational structures can more directly reveal a user’s inner reasoning, intentions, and specific objectives.
Jung Doo-won, a professor of forensic medicine at Sungkyunkwan University who has published research on generative AI forensics, explained that AI records may have stronger evidentiary value.
“While web browser searches are primarily keyword-based, interactions with AI systems are necessarily in the form of text,” says Jeong. “Prompts are written as complete statements, so the user’s actual intent is more clearly preserved.”
But experts also warn of legal and ethical concerns.
Conversations with generative AI often involve sensitive personal information, raising questions about privacy, proportionality, and the permissibility of collecting digital evidence.
Kim Myung-joo, director of the AI Safety Institute, warned against overly broad research use of AI records.
“If a crime occurs, authorities could examine the person’s entire AI conversation history and claim that criminal intent existed long before the incident,” Kim told Yonhap News. He warned that indiscriminate seizure of AI chat history could spark future human rights conflicts.
Kim also addressed the ongoing debate about AI liability, particularly when AI incites or abets a crime.
“The most difficult issue is responsibility,” he said. “For regular products, liability is governed by product liability law. AI systems don’t fit neatly into that framework. This is ultimately a challenge that society will have to solve.”
