Our Take
Courts are not yet ready to treat LLM conversations as reliable character evidence, and jurors' skepticism here is warranted: talking to a chatbot is not a crime, and venting to AI is not proof of action.
Why it matters
As prosecutors increasingly mine AI logs in criminal cases, this mistrial sets a real boundary: forensic use of chatbot interactions faces juror resistance. Defense teams and privacy advocates should watch this pattern closely.
Do this week
Legal teams: audit your AI conversation policies now. Document what you say to LLMs as if it could appear in court—because it will.
Prosecutors leaned on ChatGPT logs in arson case
Jonathan Rinderknecht faced charges for setting a fire on New Year's Day 2025 that became one of California's deadliest wildfires. Prosecutors built their case using location data from his iPhone, security camera footage, witness testimony, and his ChatGPT conversation history.
The ChatGPT evidence included requests to generate images of fire, queries like "Why am I so angry all the time?", complaints about wealthy people destroying the world, and a search asking whether someone could face blame if they lit a fire with a cigarette. Prosecutors presented these logs as part of a motive narrative.
The jury deadlocked 10-2 in favor of the defense. The judge declared a mistrial. One juror told CBS LA that the ChatGPT logs convinced her of nothing: "I talk to ChatGPT all the time." She said prosecutors' reliance on the chats actually angered her, viewing it as an attempt to criminalize routine AI use.
Jurors don't accept chatbot venting as proof of intent
This case marks an early collision between prosecutorial ambition and jury skepticism about AI evidence. Courts have not yet settled whether—or under what conditions—LLM conversations can establish character, motive, or predisposition.
The juror's comment is telling: the gap between what an AI log reveals (a person's thoughts, complaints, even troubling hypotheticals) and what it proves (the person actually committed an act) remains too wide for jurors to close on a criminal standard. Asking a chatbot questions about fire or expressing anger to an algorithm does not constitute a confession or direct evidence of arson.
Prosecutors will likely attempt this tactic again. But this result suggests juries view LLM interactions more like a diary or therapy session—a space where people explore thoughts without acting on them—rather than as reliable behavioral prediction or mens rea evidence.
What this means for your AI use
If your organization uses LLMs for brainstorming, research, venting, or exploring hypothetical scenarios, those conversations are now clearly discoverable in litigation. The Rinderknecht trial does not shield you from that discovery, but it does suggest juries will require far more than logged questions and responses to find guilt or liability.
Document your AI use policies. Train staff on what they say to public LLMs. Assume that any conversation history could be subpoenaed. And for legal and security teams: do not rely on chatbot logs as primary evidence of criminal intent without corroborating physical evidence, witness testimony, or actionable proof of planning.