Federal prosecutors say a suspect’s use of ChatGPT is now an exhibit in the case file for the Palisades Fire, joining records of calls he made and camera footage, indicating how AI chat records are becoming fair game for investigators.
Two of the devices belonging to Jonathan Rinderknecht had prompts referring to large-scale blazes and generated images of dystopian cityscapes, authorities said. Prosecutors also say he asked the chatbot whether he would be held responsible if a fire broke out from discarded cigarettes while talking to emergency operators on the phone.
Officials said the fire initially ignited with the Lachman Fire and grew into the Palisades Fire in strong winds. A dozen people were left dead, thousands of homes and businesses destroyed, and a wide array of further damage: the first toll prosecutors say was extracted by Rinderknecht’s miscalculation. Investigators also reference iPhone videos taken by the suspect, with footage of firefighters, as well as media consumed that included fires, as additional context.
What Investigators Say the Chats Reveal About Intent
Prosecutors say in court filings and briefs that the chat logs are evidentiary of intent and mindset. The prompts and images, Pontecorvo argues, reflect a lingering interest in the catastrophic burning scenes of hours past. The timing also makes a difference: some of the content, investigators say, predates the incident by months, and the question about a cigarette appears to have been asked during the time someone called 911 to report that their apartment was on fire.
Digital forensic experts say that AI chats are more like “digital journaling” of previous years, collecting words spoken or written by a person, online searches, and creative iterations. “This is a new evidentiary category,” said an expert from the SANS Institute, which deals with such problems and identified this sort of evidence as one that helps to reveal real-time thinking — whether for planning, panic, or crafting an explanation.
A New Evidentiary Frontier for AI Conversations
For years, courts have also accepted the concept of digital exhaust — browser histories in homicide and arson cases, location trails from smartphones, and even smart speaker recordings. What’s new here is the combination of text prompts and AI-generated media that are attached to a single user account and can read like a thoroughgoing narrative of ideas and intent.
Two things make AI logs super strong. First, they record iterations — multiple versions of an image or prompt that reflect evolving intent. Second, platforms usually retain metadata — when and where a post was made, what kind of device posted it — that can be used to connect activity to an individual or location. That context can be just as convincing as the content.
How Prosecutors Might Access AI Logs for Court
In order to present such material at trial against a defendant, prosecutors have to authenticate it under Rule 901 of the Federal Rules of Evidence — establish that the logs are what they purport to be. That can mean seizing a device, forensically imaging it, and comparing on-device chats to platform file records. If the provider has logs, those may come in as business records and typically meet the hearsay exception set forth in Rule 803(6).
Defense lawyers can contest the authorship, completeness, or integrity: Did a person share an account? Are timestamps consistent across sources? Whether, and to what extent, the material raises undue prejudice issues under Rule 403, particularly with graphic AI-derived images. Anticipate that courts will closely examine the chain of custody and technical documentation related to any extraction or provider processes.
Privacy, Platform Policies and the Third-Party Problem
It’s unclear in this particular case if investigators obtained the ChatGPT content from the suspect’s phone or requested it from OpenAI. Either route would raise its own familiar questions under the Stored Communications Act, which regulates law enforcement access to electronic records and usually requires a warrant for the content of communications.
The leaders of OpenAI have said in public remarks that the company holds onto some logs for a period to comply with legal and safety obligations, though users can control whether to delete chats. Civil liberties groups like the Electronic Frontier Foundation are warning that broad retention combined with the so-called third-party doctrine could mean people’s private prompts becoming subject to subpoenas and warrants, as cloud email and search history have been for years.
Recent guidance issued by the Supreme Court on digital privacy — including its decision earlier this year to extend warrant protections to historical cell-site data — indicates that courts may continue to calibrate standards for sensitive, revealing data sets. AI prompts of the kind that tend to capture personal plans and fears as well as creative hypotheticals are about right.
What’s at Stake for Average Users and Companies
The lesson for consumers is clear: Handle AI chats as you would emails or texts. They can be traceable, they may endure longer than you think, and they can offer investigators a chronology and narrative. Review the controls for data in training platforms; where possible, opt out of training, and keep in mind that deletion tools may not necessarily destroy provider-held security logs at once.
For organizations, the risk spans compliance and retention. When organizations deploy enterprise AI solutions that record prompts (to enforce policy or halt data loss), they inadvertently may be creating discoverable records. Security teams need to map where prompts live, how long they last, and who can access them in accordance with data minimization and legal hold stipulations.
The Palisades Fire case crystallizes a shift: Generative AI is not just an artistic tool — it’s an evidentiary storehouse. Amid the testing of unprecedented digital evidence by prosecutors and the weighing by courts of authenticity and privacy, AI discussion records are on track to become as much a staple of criminal investigations in the digital era as search logs and GPS traces.