Home / News In Brazil, a 36-year-old farmer was detained after he wrote to ChatGPT for several weeks about his intention to get rid of his eight-year-old son and harm other people. OpenAI provided the information to the FBI, and US authorities notified Brazilian police. The man himself denies intending to harm anyone, and charges have not yet been brought against him, writes Digital Business. The story was described by the BBC Brazilian Service. According to the publication, this is one of the rare cases when a signal from an AI developer became the reason for a police check. Now the investigation has to establish whether the chat messages were just words or whether the man actually took concrete steps. The man, whose name the BBC is withholding to protect his son, had been chatting with ChatGPT for almost two months. In the correspondence, according to police, he discussed the possibility of harming the child, other people and himself. In one of the messages, the farmer wrote that he wanted to figure out where he got the desire to cause people suffering. In another he talked about his intention to get rid of his son. It was the content of these conversations that eventually caught OpenAI's attention. The company provided information to the Federal Bureau of Investigation
US investigations, after which the FBI notified Brazilian authorities. Police in the state of Espirito Santo received information several days before the man was detained. “We gave this case priority status because of its seriousness and because the father allegedly planned to harm his son before June 20,” state police cybercrime unit chief Brenna Andrade told media. The man was detained on June 19 in a rural area of the municipality of Sao Gabriel da Paglia. During the search, law enforcement officers found four bottles with unknown substances and a rope with a tied noose. According to investigators, these items could be related to the man's plans to harm himself. Police are also looking into messages in which he allegedly discussed attacking schools, churches and law enforcement officers. Investigators believe that the man searched the Internet for information about toxic substances, including ricin, and also tried to bypass ChatGPT security restrictions by reformulating queries. “Everything that the platform reported was confirmed when the police arrived on the scene,” Andrade said. At the same time, the man himself denies that he was going to harm his son. After 54 days
The man in custody was released on August 13. No charges have been brought against him yet, the investigation continues. He is now required to wear an electronic bracelet and cannot have contact with his son or the child’s mother. Investigators are studying his phone and are waiting for the results of an examination of the seized items. In addition, they will have to find out whether the man took any specific actions to realize his intentions. Separately, his claim that he allegedly contacted a person whom he could hire to cause harm to others is being verified. The defense insists that no such actions took place. "If talking to an artificial intelligence and saying nonsense to it is a crime, then everyone will have to go to prison," the man's lawyer, Pedro Paulo Pessi, told the BBC. The story is interesting not only because of the investigation itself. It raises a much broader question: what should a chatbot do if it recognizes a potential threat to a specific person during a conversation? OpenAI told BBC Brazil that the company can notify law enforcement if it detects a "credible and immediate threat of harm to others." At the same time, investigators
received user data and messages from the company, but not ChatGPT responses. “We didn’t get the answers that were given to him,” Andrade noted. This is exactly what raises questions from the point of view of evidence. Criminal lawyer Maira Fernandez believes that the full context of the communication is needed to evaluate such messages. According to her, it is impossible to consider individual phrases without a complete history of the dialogue - it is important to understand how the conversation developed and whether the chatbot responded to the user’s statements or in any way influenced the direction of the conversation. Another expert, digital law professor Patricia Peck, believes that AI is gradually becoming not just a tool for working with information, but a system that can identify certain behavior patterns. However, she also talks about the need for clear boundaries. If the messages indicate more than just abstract considerations, but a specific potential victim, the availability of funds and an immediate risk, the technology company, it believes, may have an ethical and, in some cases, a legal duty to intervene. But turning such mechanisms into constant control over users, Peck believes, is also dangerous.
Cases of technology companies sharing information about potentially dangerous behavior of AI users with law enforcement remain rare. In August, OpenAI also provided the FBI with information about a 25-year-old Florida man after he reported intent to harm his ex-girlfriend, US media reported. Brazilian police call the farmer's case the first such case in the state of Espirito Santo and the third in the entire country where the investigation began after a signal from the AI system itself. However, in Brazil it has yet to be established whether the collected materials are sufficient to bring charges. While the investigation continues, the man remains under court-ordered restrictions.
