When talking with an AI-powered chatbot, the screen may seem like a safe space to confess and ask personal questions without fear of others' judgment. But an investigation by The Washington Post shows that if these conversations become part of a legal dispute or criminal investigation, they can quickly lose their privacy and turn from personal chats into documents and evidence used in court cases.
The Washington Post, reviewing public records and local reports, identified at least 12 cases over the past two years in which conversations with AI chatbots were used or cited.
The paper begins its review with the case of a teenage boy who turned to ChatGPT to ask about the meaning of his father's remarks about a possible financial arrangement. However, his private conversations were later made public in court records after he sued Meta, Snapchat, TikTok and YouTube, accusing those platforms of creating social media addiction and causing mental health problems.
Some of these cases resulted in settlements or dismissals. The teen's lawyers also said his ChatGPT conversation records did not affect the outcome of the case.
A record that can cause trouble
The legal value of these conversations is not limited to the questions users ask. Experts who spoke with The Washington Post say that chatting with AI bots can reveal far more information than a normal search history.
In these conversations, the user explains the background of their concerns, exchanges questions and answers with the model, and ultimately leaves a record that can reveal their thinking process, the topics they are trying to understand, and even things they may be trying to hide.
One expert described these records as "a window into the most private aspects of life."
In some cases, users' conversations are provided to police by the AI companies themselves. According to court documents cited by The Washington Post, OpenAI notified the FBI about a Florida user who repeatedly told ChatGPT he intended to attack and kill his ex-girlfriend.
The individual was later arrested, pleaded guilty, and was sentenced to eight years of probation.
OpenAI says it uses software to detect warning signs, refers related cases to human reviewers, and notifies law enforcement when a threat is judged to be imminent and likely.
Who reads users' chats?
However, this preventive approach raises another question about the boundaries of surveillance and user privacy.
The Washington Post reports that requests from government agencies and law enforcement for access to OpenAI user data are increasing rapidly. In the second half of 2025, the company disclosed information on more than 80 accounts-a number more than four times the tally for the same period the year before.
The problem, the paper notes, is that conversations with AI chatbots currently do not enjoy the same legal protections afforded to, for example, the confidentiality of communications between doctor and patient or lawyer and client.
Sam Altman, CEO of OpenAI, has called for special protections for these conversations, noting that users come to chatbots with very sensitive and private questions.
But legal experts emphasize that courts have not yet classified these communications as legally protected. In one New York case, a federal judge declined to treat a financial official's conversations with the model "Claude" as confidential, similar to lawyer–client communications, because the model is not a lawyer.
The Washington Post concludes its review with a warning beyond current debates. As AI becomes increasingly integrated into everyday life, its role in investigations and legal proceedings is expected to grow.
The investigation quotes a law professor who warns that chatbots are only "the tip of the iceberg," since more parts of users' digital worlds may come under law enforcement scrutiny in the future.