Will I get the restaurant deposit back if my spouse is arrested?
A key piece of evidence showing the reason behind Ms. A, a 50-year-old woman who wrongly accused her former husband of being a sex offender, came from ChatGPT rather than KakaoTalk messages. On June 18, the Pyeongtaek branch of the Suwon District Prosecutors' Office charged Ms. A with making false accusations, obstructing official duties through deception, and committing fraud. During the inquiry, it was discovered that Ms. A plotted the fake report with her boyfriend after her ex-husband started legal proceedings to reclaim a restaurant deposit he had once provided.
Authorities verified via digital forensics on her confiscated smartphone that Ms. A had consulted ChatGPT prior to the incident about whether her former husband could be arrested and if the deposit might be retrieved. This interaction with the artificial intelligence served as "irrefutable evidence" showing the financial motivation and deliberate planning involved in the crime.
◇Search fields capture keywords, AI understands purpose
Discussions with advanced artificial intelligence systems such as ChatGPT are becoming an increasingly common form of evidence in legal cases and inquiries. In the past, investigators mainly looked at browsing histories and KakaoTalk messages to understand a person's motivation and purpose. Nevertheless, as AI becomes more integrated into everyday activities, queries and responses generated by individuals during interactions with AI are currently being utilized as proof to uncover their thoughts and actions prior to and following criminal incidents.
Artificial intelligence chat logs tend to be more open compared to browsing history. Unlike search bars which keep just brief phrases, AI interactions record users' intentions, circumstances, and worries in full sentences. Conversations through KakaoTalk might include humor, hyperbole, or disagreements because of interaction with someone else, whereas AI queries usually ask users to explain scenarios clearly to receive accurate responses. For law enforcement, this implies they can view both what an individual looked up and the reasons behind their inquiry.
A notable example is the Gangbuk-gu motel homicide. The individual, charged with murdering two men after administering drug-laced beverages, claimed there was no intention to cause death. Nevertheless, law enforcement obtained evidence indicating the suspect had queried ChatGPT: "Could someone die from mixing sleep medication with alcohol?" The conversation revealed the suspect explicitly sought information regarding potential fatality under specific circumstances, rather than simply searching for terms like "sleeping pills" or "alcohol." Taking this into account, authorities filed murder charges, asserting the person was well-aware of the danger of dying.
Criminal defense lawyers are also concerned about artificial intelligence-generated data. Attorney Seong Jae-hwan, who previously worked as a researcher at Samsung Electronics, highlighted a recent incident related to breaches of the Information and Communications Network Act, wherein the accused used ChatGPT to request code aimed at attacking another person's website. "Although defendants might debate unrelated chats found in KakaoTalk investigations, they typically present ChatGPT logs willingly when asked," he stated. "Artificial intelligence queries frequently contain clear objectives and parameters designed to achieve certain outcomes, which can directly connect them to illegal intentions."
◇Admissions made to AI as legal proof
In international contexts, AI chat logs are increasingly being scrutinized during legal proceedings and court cases. In April, a 20-year-old individual accused of murdering two doctoral candidates at the University of South Florida had previously questioned ChatGPT about scenarios such as "What occurs if someone places another person inside a black garbage bag and discards them in a bin?" and "Is it possible to alter a car’s VIN?" Additionally, in the January 2025 fire incident in LA Palisades that resulted in 12 fatalities, evidence included AI-generated imagery created by the suspect, Jonathan Lindnerknecht, showing scenes of urban flames.
This month, the U.S. District Court within the Southern District of New York approved the implementation of a search warrant targeting the chat history from an OpenAI bot involving Richard Kim, a previous official at a crypto firm facing charges of securities fraud. Kim's team claimed these records could contain discussions pertinent to their legal strategy; however, the court determined that interactions with an artificial intelligence chatbot do not inherently qualify for attorney-client confidentiality.
Sam Altman, the head of OpenAI, has earlier advocated for "confidentiality protections for AI conversations." He expressed worries that if every discussion with ChatGPT were made available to law enforcement or courts, it might violate personal privacy since individuals utilize ChatGPT not only as an information source but also as a trusted companion or journal. Nevertheless, according to present regulations, artificial intelligence isn’t considered a legal professional, physician, or therapist; therefore, the content shared with AI does not come with automatic legal protection.
There are growing demands in South Korea to set guidelines for utilizing AI-generated chat transcripts as evidence during investigations. Although such data can provide valuable insights into a suspect's intentions and motivations, they frequently include irrelevant personal discussions and confidential business information.
Professor Jeong Du-won from the Department of Forensic Science at Sungkyunkwan University stated, "Conversational AI functions as a search engine, messaging platform, and productivity tool. Although direct inquiries made during planning stages of crimes may readily expose illegal intentions, personal and professional communications often overlap. It is necessary to conduct distinct analyses based on current guidelines for message forensic examinations."