Portuguese 
Spanish 
Conversations mentioned victim, weapons, poison, and a date for the attack, leading Brazilian authorities to arrest the suspect before the possible crime.
The child was only 8 years old. Furthermore, according to the messages, the crime was to be committed the day after the arrest.
Given the seriousness, OpenAI, the company responsible for ChatGPT, sent an alert to the FBI.
-
Scientists Discover 125-Million-Year-Old “Pregnant Fossil” with Preserved Embryos and Microscopic Larvae Inside Mollusk in England
-
17-Year-Old Justin Bernstein Develops Bacteria to Produce Glass on Arctic Ice to Reflect Sunlight, Wins $50,000 U.S. Scholarship
-
The 19-Meter Rocket-Helicopter: A Private Industry Prototype That Challenged Space Race Logic
-
Brazilian Students Convert Seized Illegal TV Boxes into Linux-Powered Mini Computers for Schools and Social Projects Across São Paulo
Subsequently, the American agency forwarded the information to the Ministry of Justice and Public Security, which contacted the Civil Police of Espírito Santo.
Thus, the authorities were able to locate and arrest the suspect before the date mentioned in the conversations.
Messages presented details of the possible crime
During the conversations, the man wrote that he would have a weapon, a rope, and poison.
Additionally, he mentioned the intention to carry out attacks in public spaces.
The suspect also stated that he tried to hire a hitman for R$ 50,000 to kill his son.
However, according to the account, the proposal was refused after the potential hire discovered that the victim would be a child.
In another message, the man questioned the origin of his desire to kill people. At the same time, he declared that he enjoyed watching others suffer.
Therefore, the conversations contained elements considered serious, such as identified victim, methods, instruments, and a near date.
How ChatGPT identifies threats
OpenAI informed g1 that its systems were developed to protect users and others.
Firstly, the analysis is carried out by automated systems, which can classify content and interpret the context of messages.
Additionally, the process uses lists of prohibited terms and databases with previously flagged materials.
When a conversation shows signs of risk, it can be forwarded to human moderators.
Then, these professionals analyze whether there was a violation of policies and verify the real possibility of a violent act.
If the risk is considered low, no extreme measures need to be taken.
However, in the face of an imminent and credible threat, the account may be deactivated. Additionally, competent authorities may be notified.
Conversation pattern increased the severity
According to Álvaro Machado Dias, a professor at the Federal University of São Paulo, digital assistants analyze content mainly through automatic mechanisms.
However, the case in Espírito Santo exceeded the risk limit because there was a victim, available means, and a defined date.
Therefore, according to the expert, the alert reached the maximum level of severity.
In a statement published in April 2026, OpenAI stated that certain dangers appear only after analyzing longer conversations.
In other words, a single message may seem harmless. However, the set of interactions can reveal more concerning behavior.
International alert reached Brazilian police
After receiving the communication from OpenAI, the FBI sent the material to the Ministry of Justice and Public Security.
Subsequently, the information was forwarded to the Specialized Cybercrime Repression Police Station of Espírito Santo.
The team executed judicial warrants and arrested the man on June 19.
According to delegate Ícaro Olímpio, the suspect denied the accusations.
However, the conversation history provided by OpenAI presented enough elements to justify the preventive action.
According to delegate Breno Andrade, in an interview with TV Gazeta, this was only the third similar alert received in Brazil.
Expert advocates for direct channel in Brazil
Although he considered the communication of the case correct, Álvaro Machado Dias criticized the initial sending to the FBI.
According to him, the alert should have been sent directly to the Federal Police or the Ministry of Justice.
Even so, the expert acknowledged the logic of OpenAI, which is based in the United States and has contact with American authorities.
At the same time, he advocated for the creation of a specific Brazilian channel to receive reports produced by artificial intelligence platforms.
In this way, investigations could start more quickly in risk situations.
Finally, the professor emphasized that conversations with ChatGPT do not have professional secrecy.
Therefore, although some people treat artificial intelligence as a therapist or lawyer, the system does not perform any of these functions.
Do you believe that artificial intelligence companies should immediately report conversations that indicate real threats against other people to the authorities? Share your opinion!
