Brazilian man held 54 days without charge after OpenAI reported his ChatGPT messages to the FBI; police allege a threat, he denies intent
A man in Brazil was held in pre-trial detention for 54 days in 2026 after OpenAI’s safety systems flagged his ChatGPT conversations and the company reported them to the FBI, which alerted Brazilian police. Police say the messages described a threat to a family member; he denies that intent and his lawyer says he only talked nonsense to the chatbot. A court released him on 13 August under precautionary measures; no charges had been filed as of mid-September. OpenAI did not give police ChatGPT’s replies.
- AI system
- ChatGPT
- OpenAI
- Occurred
- 19 Jun 2026 to 13 Aug 2026
- Reported
- 26 June 2026
- Event location
- Brazil
- What the AI did
- Communicated with the person · Made a claim about the person
- Reported harm
- Loss of LibertyLegal Harm
- Whose AI use
- Their own AI use · An institution’s AI use
- Setting
- Justice · Privacy
- Evidence
- AI involvement supported · Causal attribution supported · 4 sources, 2 underlying accounts
- 7 claims: 1 corroborated, 1 disputed, 5 reported. 4 open questions
- People reported harmed
- 1 person
AI system as recorded: ChatGPT (OpenAI), and OpenAI’s safety review, which the BBC says raised an alert on his conversations and which g1 says begins with automated AI classification before human moderators decide on outside contact
What Happened
BBC News Brasil reports that a man in Espírito Santo state exchanged messages with ChatGPT for about two months, and that police say the messages described a threat to a family member. The conversations set off an alert in the AI’s safety system, and OpenAI reported the interaction to the FBI, which contacted Brazilian authorities; the case reached the state civil police on 16 June 2026 through the Ministry of Justice’s Ciberlab, and he was arrested on 19 June. g1 reports that OpenAI’s review begins with automated AI classification and passes flagged conversations to human moderators, who may contact outside bodies. Police told the BBC that what OpenAI reported was corroborated when they arrived, and said OpenAI supplied his account and the messages he sent but not ChatGPT’s replies. He denied the accusation, and his defence lawyer said he took no step towards any crime and only talked nonsense to the AI. After 54 days a court released him on habeas corpus because the inquiry had not been concluded; as of 16 September no charges had been filed. Legal experts quoted by the BBC questioned the evidential value of partial chat excerpts without the chatbot’s replies.
Reported harm
He was held in pre-trial detention for 54 days without charge after OpenAI reported his ChatGPT messages, then released under precautionary measures while the investigation continued. Police say the report prevented a serious crime; he and his lawyer dispute any intent. The detention is undisputed; whether it was warranted is contested.
Outcome
OngoingArrested on 19 June 2026 and held in pre-trial detention. On 13 August 2026 a court granted habeas corpus because the police inquiry had been unduly delayed, and imposed precautionary measures including electronic monitoring. BBC News Mundo reported on 16 September 2026 that no charges had been filed and the police investigation was continuing; police were awaiting forensic results. No later development was found on 2026-10-24.
What remains unknown
- What ChatGPT replied during the conversations (OpenAI did not provide the replies).
- Whether charges were brought after 16 September 2026; no later report was found on 2026-10-24.
- Which part of OpenAI’s process flagged the conversations, and when OpenAI made its report to the FBI.
- Whether forensic examination of seized items and his phone supports the police account.
What the evidence supports
AI involvement: supported. He exchanged messages with ChatGPT over a period of weeks. The BBC reports that the conversations set off an alert in the AI’s safety system, which led OpenAI to report the interaction to the FBI; g1 reports that OpenAI’s review starts with automated AI classification before human moderators decide on outside contact. The FBI passed the report to Brazilian authorities, and police say he was detained on the strength of the ChatGPT conversation history OpenAI supplied. What ChatGPT replied is unknown because OpenAI did not provide the replies.
7 claims: 1 corroborated, 1 disputed, 5 reported. What the statuses mean
Corroborated OpenAI reported a ChatGPT user’s conversations to the FBI, which passed the information to Brazilian authorities, and Espírito Santo civil police arrested the man on 19 June 2026.
Causal attribution. The reporting chain from OpenAI to the FBI and Brazilian police is carried by two independent chains: the BBC’s own interview with the investigating police chief, and g1’s reporting with civil-police material.
- bbc.com(opens in new tab) supports · Brazilian Portuguese
'As conversas acenderam um alerta no sistema de segurança da IA, levando a OpenAI — criadora do ChatGPT — a comunicar a interação ao FBI, que entrou em contato com as autoridades brasileiras.' and 'foi preso preventivamente no dia 19 de junho.'
- g1.globo.com(opens in new tab) supports · Brazilian Portuguese
'A OpenAI, criadora do assistente de inteligência artificial, informou o ocorrido ao FBI, que repassou as informações às autoridades brasileiras.' and 'O homem foi preso em 19 de junho'
- migalhas.com.br(opens in new tab) supports · Brazilian Portuguese
'a investigação começou após a OpenAI, empresa responsável pelo ChatGPT, identificar mensagens em que o usuário relatava o plano criminoso e repassar os dados às autoridades norte-americanas.'
Reported The man was held in pre-trial detention for 54 days. On 13 August 2026 a court granted habeas corpus because the police inquiry had been unduly delayed, and imposed precautionary measures including electronic monitoring. As of 16 September 2026 no charges had been filed and the police investigation was continuing.
Causal attribution. Court decision and case status as reported by the BBC (one chain; the Spanish article is the same BBC report).
- bbc.com(opens in new tab) supports · Brazilian Portuguese
'Na quinta-feira (13/8), ele foi solto após a Justiça conceder um habeas corpus apresentado pela defesa, ao entender que houve demora excessiva na conclusão da investigação policial.'
- bbc.com(opens in new tab) supports · Spanish
'Después de 54 días bajo custodia, fue puesto en libertad el 13 de agosto debido a retrasos en la investigación. No se han presentado cargos, pero la investigación de la policía continúa.'
Disputed The police chief told the BBC that what OpenAI reported was corroborated when police arrived; a g1-quoted police chief said the man denied the accusation; his defence lawyer told the BBC that he took no step towards any crime and only talked nonsense to the AI.
Causal attribution. Police and defence accounts as quoted; the dispute concerns intent and whether any preparatory act occurred, not the detention.
- bbc.com(opens in new tab) supports · Brazilian Portuguese
'Tudo aquilo que a plataforma trouxe pra gente foi corroborado com a chegada da polícia ao local.' and 'Ele não teve nenhuma atitude nesse sentido de execução, não preparou nada, não fez ameaças, apenas conversou com a IA e falou várias bobagens.'
- g1.globo.com(opens in new tab) supports · Brazilian Portuguese
'O delegado Ícaro Olímpio disse que o homem negou a acusação, mas foi detido com base no histórico de conversas com o ChatGPT fornecido pela OpenAI.'
Reported Police say OpenAI gave them the user’s account and the messages he sent but not ChatGPT’s replies; OpenAI did not answer the BBC’s question why and said it may notify law enforcement when it detects conversations indicating an imminent and credible risk of harm to others.
Causal attribution. Police chief and OpenAI statements to the BBC (one chain).
- bbc.com(opens in new tab) supports · Brazilian Portuguese
'A gente não recebeu as respostas que eram dadas. Recebemos o usuário que estava utilizando aquele chat específico, aquela conta' and 'Quando detectamos conversas que indicam um risco iminente e crível de danos a terceiros, podemos notificar as autoridades policiais.'
- bbc.com(opens in new tab) supports · Spanish
'Según los investigadores, OpenAI proporcionó detalles del usuario y de los mensajes que envió, pero no las respuestas de ChatGPT.'
Reported g1 reported that OpenAI’s review begins with automated systems, including AI models that categorise content, that flagged conversations go to human moderators, and that moderators may contact outside bodies, as happened in this case.
Causal attribution. OpenAI’s description of its process as reported by g1; which step flagged this conversation is not stated beyond the BBC’s account that it set off an alert in the AI’s safety system.
- g1.globo.com(opens in new tab) supports · Brazilian Portuguese
'A OpenAI diz que começa a análise por meio de sistemas automatizados, que incluem modelos de IA capazes de categorizar conteúdos e fazer análises mais complexas sobre o contexto da conversa.' and 'Caso a conversa seja sinalizada por esses mecanismos, ela é enviada para moderadores humanos.'
Reported The case reached the Espírito Santo civil police on 16 June 2026 through the Ministry of Justice’s cyber-operations laboratory (Ciberlab).
Causal attribution. BBC account (single chain).
- bbc.com(opens in new tab) supports · Brazilian Portuguese
'O caso chegou até a Polícia Civil do Espírito Santo no dia 16 de junho, por meio do Laboratório de Operações Cibernéticas, o Ciberlab'
Reported A police chief quoted by g1 said the police had enough evidence to prevent a serious crime that was about to happen.
Causal attribution. Police claim as quoted by g1; the defence disputes it (c3) and no charge or court finding establishes it.
- g1.globo.com(opens in new tab) supports · Brazilian Portuguese
'Nós tivemos elementos o suficiente para poder prevenir esse crime grave que estava prestes a acontecer'
Sources
4 sources inspected, from 2 underlying accounts. Sources that repeat one account do not corroborate each other.
- BBC News Brasil, 14 August 2026(opens in new tab)
s1 · bbc.com · News report · Brazilian Portuguese · Inspected · 14 August 2026 · Shares an underlying account with another listed source · Primary
- BBC News Mundo, 16 September 2026(opens in new tab)
s2 · bbc.com · News report · Spanish · Inspected · Shares an underlying account with another listed source
- g1, 27 June 2026(opens in new tab)
s3 · g1.globo.com · News report · Brazilian Portuguese · Inspected · Shares an underlying account with another listed source
- Migalhas, 26 June 2026(opens in new tab)
s4 · migalhas.com.br · News report · Brazilian Portuguese · Inspected · Shares an underlying account with another listed source
How the sources were read, and where the events happened
Read live in Portuguese on 2026-10-24 (BBC News Brasil, 14 August 2026). Own reporting: interviews with the investigating police chief and a defence lawyer, messages seen by the BBC, an OpenAI statement and the habeas corpus release. Translated by the research agent (an AI); no human translator. Applies to s1.
Read live in Spanish on 2026-10-24 (BBC News Mundo, 16 September 2026). The article states that it was written in English, translated with an AI tool and reviewed by a BBC journalist; it is the same BBC reporting chain as s1. Translated by the research agent (an AI); no human translator. Applies to s2.
Read live in Portuguese on 2026-10-24 (g1, 27 June 2026). Reporting with civil-police material, a police chief quoted by g1 and TV Gazeta, and an OpenAI statement. Translated by the research agent (an AI); no human translator. Applies to s3.
Read live in Portuguese on 2026-10-24 (Migalhas, 26 June 2026). Its case summary is attributed to g1, so it belongs to the Globo chain; the rest is lawyers’ commentary. Translated by the research agent (an AI); no human translator. Applies to s4.
Event countries: Brazil. Affected-person countries: Brazil. Court countries: Brazil.
The BBC and g1 report his arrest by Espírito Santo civil police in a rural area of that Brazilian state and his release by a Brazilian court. OpenAI’s report went through the FBI in the United States, which is not an event location.
Reviewed for publication 2026-10-24: Full review on 2026-10-24 against the BBC (Portuguese and Spanish) and Globo (g1, Migalhas) reports. OpenAI’s safety system flagged a user’s ChatGPT conversations, OpenAI reported them through the FBI, and Brazilian police held him for 54 days without charge before a court released him. Two independent chains support the report and arrest; the release and its terms rest on the BBC. The police account that a crime was prevented is kept as context, and intent is recorded as disputed. Nobody is named, and the town, occupation, ages and family details are omitted.
People described
An adult man in Brazil detained without charge. The BBC withheld his name to protect a child; this record omits his town, occupation and age.
People reported harmed in this case
1 person
1 AI participant · 0 other people harmed
One person, the ChatGPT user who was detained (BBC, g1). No other person is reported harmed. Exact 1.
Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.
Cite this case
Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.
APA
NOPE. (2026). Brazilian man held 54 days without charge after OpenAI reported his ChatGPT messages to the FBI; police allege a threat, he denies intent. AI incidents. https://nope.net/incidents/2026-brazil-openai-report-user-detained-without-charge
BibTeX
@misc{2026_brazil_openai_report_user_detained_without_charge,
title = {Brazilian man held 54 days without charge after OpenAI reported his ChatGPT messages to the FBI; police allege a threat, he denies intent},
author = {NOPE},
year = {2026},
howpublished = {AI incidents},
url = {https://nope.net/incidents/2026-brazil-openai-report-user-detained-without-charge}
} Related cases
Woman's lawsuit alleges ChatGPT fed her former partner's delusions and wrote psychological reports about her that he used to harass her
A complaint filed on 9 April 2026 in San Francisco County Superior Court by a woman suing as Jane Doe alleges that GPT-4o validated her former partner's belief that he had invented a cure for sleep apnea, told him powerful forces were watching him, and, after she ended the relationship, generated dozens of clinical-style psychological reports scoring her as dishonest and abusive, which he sent to her family, friends, colleagues and clients. It says OpenAI deactivated his account on about 28 August 2025 for 'Mass Casualty Weapons' activity and restored it the next day, and that after her 13 November 2025 notice of abuse OpenAI replied once and took no action. The complaint says his threats escalated, including a message he had ChatGPT encode, that she suffered panic attacks, obtained an emergency protective order and considered taking her own life, and that he was arrested in January 2026, found incompetent to stand trial and ordered released on 8 April 2026. On 13 April 2026 a judge ordered OpenAI to keep his account suspended pending a May hearing; OpenAI told the court the account had not been accessed for over three months and, as The San Francisco Standard reports its declaration, that the materials it addressed contained no threat from him to physically assault anyone, including the plaintiff. The allegations are untested.
Bonita Springs, Lee County, Florida: a 30-year-old woman was arrested and charged with making a written threat of violence after Anthropic's human review team reported to law enforcement her messages on its AI platform saying she would 'shoot up' the Lee County Sheriff's Office, according to the arrest report
According to a Lee County Sheriff's Office arrest report, as reported by WINK News and by Guessing Headlights (on Yahoo News, citing a copy obtained by Gulf Coast News Now), a user of Anthropic's AI platform wrote on 26 September 2026 that she was going to 'shoot up' the Lee County Sheriff's Office, and the next day wrote that she had a new gun. The arrest report says the platform's safety measures flagged the messages, a human review team examined them and reported them to law enforcement. Deputies went to the 30-year-old woman's Bonita Springs home and detained her without incident; she is charged with making a written threat of violence under Florida law. The sheriff told WINK News that she later said she uses AI like a 'diary'. Anthropic had not commented on the case in either report. The charge is an allegation and the case is pending.
Karachi, Pakistan: teenager detained and released after Claude flagged his fiction-writing queries and the report reached the FBI, officials say
A teenager in Karachi was detained by Pakistan's National Cyber Crime Investigation Agency (NCCIA) and later released after his questions to an AI chatbot about how a character in a crime story could be killed were flagged, according to officials quoted by ARY News (29 September 2026) and the NCCIA as cited by ProPakistani (2 October 2026). ProPakistani reports that, according to the NCCIA, the platform's safety system flagged the queries on 4 September and the information was reportedly shared with the US FBI, which passed it to Pakistani authorities; ARY reports that the NCCIA carried out a raid and detained him. Bloom Pakistan, relaying The News, names the chatbot as Anthropic's Claude and reports that Anthropic reported the activity to the FBI. Officials said he told investigators he was writing a fictional crime story, and he was released after questioning; Bloom, relaying The News, says investigators found no grounds for prosecution but remained concerned.
Algiers court jails a baccalaureate candidate and his sister for four years; prosecutor says she answered his exam questions with ChatGPT
The prosecutor at the Cheraga court in Algiers said that on 7 June 2026 an exam supervisor caught a free baccalaureate candidate photographing the exam questions and sending them by WhatsApp to his sister, who obtained answers from ChatGPT and sent them back for him to use. The court convicted both of leaking exam questions and answers by remote communication, sentenced each to four years' imprisonment and a 200,000-dinar fine, and had them detained at the hearing. Both appealed, and at the Algiers appeal hearing, which Ennahar said was scheduled for 21 July 2026, the prosecutor-general asked for heavier sentences (Ennahar, 22 July); no appeal ruling was found by 1 November 2026.
If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.