The Arson Case That Used ChatGPT Logs as Evidence: A Terrifying New Legal Precedent

Last week, I found myself staring at a screenshot of someone's ChatGPT conversation, and I couldn't shake the feeling that I was looking at something deeply unsettling. Not because the conversation was particularly dark—it was about how to fix a broken sprinkler system—but because of where I saw it: on a news site, attached to a criminal case.

Here's the thing: we've gotten used to the idea that our texts, emails, and search histories can be used against us in court. That's been true for years. But what happens when the thing you're talking to isn't a person, but a large language model? What happens when your private conversations with an AI become evidence in a murder trial?

We just got our first real-world answer.

The Fire That Changed Everything

On New Year's Day 2025, Jonathan Rinderknecht allegedly set a fire that would become one of the deadliest in Los Angeles history. The Palisades fire tore through the Santa Monica Mountains, destroying homes, displacing thousands, and killing at least a dozen people. It was the kind of tragedy that makes you stop and think about how fragile everything is.

Prosecutors built a case using the usual tools: iPhone location data, security camera footage, and witness testimony. But they also had something I've never seen used in a criminal trial before: logs from Rinderknecht's ChatGPT sessions.

According to www.theverge.com, the prosecution argued that Rinderknecht had used ChatGPT to "research how to start a fire, how to avoid detection, and what charges he might face." The logs showed a progression from curiosity to planning to, allegedly, action. It's the kind of digital breadcrumb trail that investigators have been dreaming about for years.

The New Digital Fingerprint

Let's be honest: we all know that anything we type into a search engine or a messaging app could theoretically be subpoenaed. But there's something different about ChatGPT. When you talk to a language model, you're not just searching for information—you're having a conversation. You're thinking out loud. You're asking questions you might never ask another human being because you assume the AI doesn't judge you and won't tell anyone.

That assumption is now officially wrong.

The logs in question reportedly showed Rinderknecht asking about specific accelerants, the best times to set fires in dry brush, and even asking the AI to role-play as an arson investigator to see how he might be caught. It's the kind of thing that, in a vacuum, looks absolutely damning. But here's where it gets complicated: was he planning a crime, or was he just morbidly curious? Was he researching out of genuine intent, or was he like the millions of people who ask ChatGPT dark hypotheticals because they find it fascinating?

I've personally asked ChatGPT how to pick a lock—not because I wanted to break into anything, but because I was writing a scene in a short story. My search history includes questions about poison detection, surveillance techniques, and how to disappear. If those became evidence, I'd look guilty as hell.

The Mistrial and the Messy Questions

Here's where the story gets even weirder. The trial ended in a mistrial. Not because the ChatGPT evidence was excluded, but because of a juror who apparently conducted their own independent research. The jury was deadlocked, and the judge declared a mistrial.

But the damage—or the precedent, depending on how you see it—is already done. According to www.theverge.com, the prosecution's use of ChatGPT logs was "a novel approach that could set a precedent for how AI conversations are treated in court." The defense argued that the logs were protected by privacy laws and that using them violated Rinderknecht's rights. The judge allowed them in.

That's the part that keeps me up at night. Not because Rinderknecht is innocent—I have no idea if he is—but because the legal system just decided that your conversations with an AI are fair game. And unlike a phone call or an email, there's no expectation of privacy with ChatGPT. OpenAI's privacy policy explicitly states that they may share data with law enforcement under certain circumstances. Most of us clicked "agree" without reading it.

What This Means for the Rest of Us

I spend about three hours a day talking to ChatGPT. I use it for work, for creative brainstorming, for therapy-adjacent venting sessions. I've asked it about my deepest fears, my most embarrassing moments, and things I would never tell another living soul. I assumed—naively, I now realize—that those conversations existed in a kind of gray area. Not quite private, but not quite public either.

This case changes that calculation.

If you're using ChatGPT or any other AI chatbot, here's what you need to know: everything you type is being logged. Those logs can be subpoenaed. They can be used against you in court. And unlike a private conversation with a lawyer or a therapist, there's no privilege protecting them.

Prosecutors are already salivating at this new tool. Imagine a domestic violence case where the abuser asked ChatGPT for manipulation tactics. A fraud case where someone asked the AI how to forge documents. A murder case where someone asked about disposing of a body. The AI becomes a perfect witness—it never forgets, never lies, and never refuses to testify.

The Other Side of the Coin

Let me be clear: I'm not saying the prosecution was wrong to use the logs. If Rinderknecht did what he's accused of, those logs are powerful evidence. They show intent, planning, and premeditation in a way that circumstantial evidence often can't. In some ways, it's the cleanest evidence you could hope for.

But here's the problem: we don't have clear rules about this yet. There's no Supreme Court ruling on what constitutes a reasonable expectation of privacy in an AI conversation. There's no precedent for whether asking a hypothetical question to an AI is the same as planning a crime. We're in a legal gray zone, and the first case to test it ended in a mistrial.

I talked to a lawyer friend about this last week. She told me that the real issue isn't whether the logs are admissible—it's whether people understand that they can be used. "Most of my clients," she said, "think that if they delete something, it's gone. They don't understand that the service provider still has it. And they definitely don't understand that an AI model might be recording their questions."

What Comes Next

The Palisades fire trial is scheduled for a retrial. The prosecution will almost certainly use the ChatGPT logs again. The defense will almost certainly appeal if there's a conviction. Eventually, this case—or one like it—will reach a higher court, and we'll get some clarity.

But until then, we're all living in a world where your late-night conversations with an AI about your darkest thoughts could end up on a prosecutor's screen. I'm not telling you to stop using ChatGPT. I'm not even telling you to be more careful about what you ask. I'm just telling you that the old rules don't apply anymore, and we don't know what the new rules are.

Honestly? That scares me more than the fire itself.

---

Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by Michael Reeves.