The Palisades Fire is one of the worst urban wildfires in American history. It burned more than 23,000 acres, destroyed nearly 7,000 structures, and killed 12 people in Los Angeles in January 2025. The man accused of starting it, Jonathan Rinderknecht, just had his first federal trial end in a mistrial, with 10 of 12 jurors voting not guilty. A retrial is already scheduled for October 19.
What made this trial notable, beyond the scale of the disaster, was how prosecutors built part of their case. Alongside location data from his phone and security camera footage, they presented his ChatGPT conversation history. They pointed to AI-generated images of fire he had requested, questions he asked the chatbot about anger, and a recorded screen session where he asked whether someone could be blamed for a fire started by their cigarette.
One juror told reporters she was unimpressed. She said the ChatGPT logs showed nothing and that she uses the chatbot herself regularly. It made her angry that the prosecution was framing ordinary chatbot use as a character flaw. That reaction is genuinely interesting: jurors are now ChatGPT users too, and they have a personal frame of reference for what it means to vent to an AI.
But the jury's reaction does not change the underlying legal reality, and that is the part worth paying attention to.
Courts treat conversations with AI chatbots exactly the same as phone records, emails, or credit card swipes. There is no legal protection equivalent to the confidentiality of a lawyer, therapist, or doctor. OpenAI's own CEO has publicly acknowledged this gap. A cybersecurity attorney and legal expert described AI chat logs to CNN as "a treasure trove for law enforcement" because people wrongly assume the conversations stay private.
This has already moved well beyond one arson trial. In a copyright case in late 2025, a federal judge ordered OpenAI to hand over 20 million user conversation logs as evidence. The users whose conversations were included were not told and had no way to object. Courts have also confirmed that even if a user deletes a chat, OpenAI may still hold it server-side, and a court order can freeze deletion entirely.
In another recent criminal case, a suspect's ChatGPT questions, including asking whether he could legally keep a gun at home and how to change a car's identification number, appeared in a criminal affidavit. The Palisades fire case is part of a clear pattern, not a one-off.
For professionals who use AI tools at work, the practical picture is this: anything typed into a public AI chatbot is a record. It can be subpoenaed in a lawsuit, requested by regulators, or obtained by law enforcement. If an employee pastes confidential business information, drafts a sensitive HR decision, or works through a legal question using a consumer AI tool, that conversation exists and can surface later.
The distinction courts are starting to draw is between enterprise AI contracts, which often include specific data handling and retention terms, and standard consumer or free-tier accounts, which follow OpenAI's general retention policies. If your organisation has staff using AI tools, the type of account and the contract terms around data retention are worth checking.
The Rinderknecht retrial starts in October. Whether or not the ChatGPT logs eventually help secure a conviction, they have already established something broader: AI conversation history is now ordinary courtroom material, and most people using these tools have no idea.