In the federal arson trial over the January 2025 Palisades fire in Los Angeles, prosecutors built part of their case on the defendant's ChatGPT logs. The jury's 10 2 split, and one juror's blunt reaction, suggests ordinary AI chatbot use is not yet treated like other evidence.
A Los Angeles-area federal jury couldn't agree on whether Jonathan Rinderknecht started the Palisades fire, the deadly January 2025 Los Angeles-area wildfire. But they were unusually clear about one thing: prosecutors' decision to use his ChatGPT logs as evidence did not sit well.
"I talk to ChatGPT all the time," said one juror who spoke to CBS LA, in remarks relayed by The Verge. She said the use of those logs in the case made her "angry." On Friday, U.S. District Judge Percy Anderson declared a mistrial after the jury deadlocked 10-2 in favor of acquittal, according to the Los Angeles Times and ABC News.
The split is not an acquittal. Prosecutors can retry the case, and the mistrial leaves no binding precedent. But it does leave a question on the record: when prosecutors offer a defendant's ordinary conversations with an AI chatbot as proof of motive, what are jurors supposed to do with them?
This appears to be the first prominent federal criminal trial in which ChatGPT logs were treated as substantive evidence of intent, according to The Verge's reporting on the case. The prosecution's theory, as described in trial coverage by PBS NewsHour and NBC News, was that Rinderknecht used OpenAI's ChatGPT to generate images of fire and to vent grievances about wealthy people in the months before the New Year's Day 2025 blaze. Prosecutors pointed to a prompt in which he allegedly asked "Why am I so angry all the time?" and to rants in which he accused the wealthy of destroying the world. They also entered a screen recording in which, they said, he asked whether someone could be blamed for a fire started by their cigarette.
The defense's response was practical. The ChatGPT outputs were generated text, not confessions, and a modern jury uses these tools daily. The juror's remark to CBS LA captured the defense theory in plain English: this is how people talk to chatbots now.
That reaction is the more durable piece of the story. Federal evidence rules are built around documents, sworn testimony, physical evidence, and expert interpretation. ChatGPT logs do not fit cleanly into any of those categories. They are conversational artifacts. They can be prompted into existence by anyone with an account, they reflect what a model produced in response to a query, and they are easy to misread without context. None of those properties is, on its own, a reason to exclude them. But they also do not behave like a diary, a recording, or a search history, and courts have not yet settled how to treat them.
Jury skepticism toward AI-derived evidence is not new, but it has usually been a hypothetical. The Palisades case put ordinary jurors in the position of weighing personal chatbot logs alongside iPhone location data, security camera footage, and witness testimony, and a meaningful share of them did not find the logs persuasive. That is a signal worth watching, not a rule.
For now, the practical questions are small and concrete. Will prosecutors retry Rinderknecht, and if so, will they again lean on ChatGPT logs as a centerpiece of the case? Will defense lawyers in other matters now move to exclude similar logs, on the grounds that they confuse jurors more than they clarify? And as AI chatbot use becomes ordinary across the population, will courts treat a defendant's prompts to a large language model as closer to a diary entry, a brainstorm, or something the law has not yet named?
The Palisades fire killed people, and the legal process around it is not over. The mistrial bought time for both sides, and it bought the broader question of AI-as-evidence another round of public attention it had not yet had.