ChatGPT can turn you in to the police: which conversations are dangerous

Conversations with ChatGPT do not guarantee confidentiality. OpenAI may share data with police if a credible threat of serious harm is detected. A case is described where correspondence led to the arrest of a man planning violence. Law enforcement can also access correspondence through the user's device. The company receives official requests from government agencies.

Millions of people use ChatGPT to discuss personal issues, but conversations with AI do not have the same legal protection as communication with a lawyer. OpenAI may share information with law enforcement if automated systems or specialists identify a credible and immediate threat of causing serious harm to others. The decision depends on the nature of the threat and the circumstances. A real case from the US is cited: a man discussed with ChatGPT his intention to commit sexual assault and murder of his ex-girlfriend. OpenAI shared the data with police, the man was arrested, pleaded guilty, and received eight years of probation. Another case is described where a student who damaged cars asked ChatGPT about the likelihood of being caught; police found the correspondence on his phone. OpenAI receives official requests from government agencies: in the second half of 2025, there were 75 requests for data content and 10 emergency requests. Users are advised not to share passwords, bank details, or confidential third-party data with AI. Deleted chats are stored for up to 30 days, and temporary chats do not guarantee anonymity.