The man accused of setting the massive Palisades Fire in Los Angeles repeatedly asked ChatGPT to create images of a city engulfed in flames and floods, and cursed at OpenAI’s chatbot when it told him the scenery he envisioned was too violent, jurors at his trial were told on June 11. Months before Jonathan Rinderknecht allegedly set the Jan 1, 2025, brush fire that the authorities say later morphed into the blaze that destroyed one of the city’s richest neighbourhoods, he repeatedly instructed the chatbot
to create images of wealth inequality featuring mass devastation, according to a federal agent who helped retrieve data from his two phones. Jurors saw a transcript of Rinderknecht’s exchange with ChatGPT showing that the chatbot initially rejected his requests, telling him the violent scenes were against the company’s policy. Prosecutors argue that Rinderknecht was upset that he was alone on New Year’s Eve and was motivated by his anger at social and financial inequality to set a fire in a state park atop the Pacific
Palisades with a barbeque lighter. Rinderknecht’s lawyer contends he is innocent, and has been made a “scapegoat” for the failure of firefighters to prevent the quickly contained Jan 1 Lachman Fire from rekindling six days later into the massive Palisades Fire after hurricane-force winds stoked underground embers. In one prompt, Rinderknecht, then 29, asked ChatGPT for a depiction of a tsunami of water as well as “a bunch of trees burning and the first part of the population is running away and some of them
are looking back and screaming and suffering and drowning”. ChatGPT responded: “The specific request still includes elements that are not compliant with our content policy, such as the depiction of widespread suffering.” “That’s the reality,” Rinderknecht replied, according to the transcript. Jurors also saw the defendant berate and curse at the chatbot after it refused to comply. “No, what are you f**cking doing?” he said. “Stick to what you said initially. On the left a scene showing environmental destruction.” US District Judge Anne Hwang, who
is presiding over the trial, barred the government from showing the jury the artificial intelligence imagery after Haney argued it would be “very, very prejudicial”. But prosecutors were permitted to display the prompts Rinderknecht gave to the chatbot to generate images he wanted. Michael Montevidoni, an agent of the Bureau of Alcohol, Tobacco, Firearms and Explosives, testified that when the authorities questioned Rinderknecht about three weeks after the Lachman Fire, he refused to provide the passwords to his phone, telling them at one point that
“it would take thousands of years” to get into his iPhone. The agent said Rinderknecht employed a Tor browser to access the “dark web,” which made his phone “completely undetectable,” and also installed software on his computers “that would remotely wipe the phones automatically”. Investigators reviewed Rinderknecht’s social media accounts and found that he frequently focused on wealth inequality and climate change, Montevidoni said. “They were showing his anger and frustration,” the agent told jurors. One request to ChatGPT asked the chatbot to generate an
image with Mother Nature on one side and corporations on the other, Montevidoni testified. “He talked about the dark side of humanity and how they’re enslaving the population and destroying the environment,” the agent said. OpenAI did not immediately respond to a request for comment. Rinderknecht’s Google search history showed phrases such as “kill all the billionaires” and “fires burning against a wealthy city,” Montevidoni said. Investigators were also able to develop other evidence tied to his phones. His phone data usage placed him on
the hillside where the Lachman Fire started shortly after midnight on Jan 1, 2025, and security camera footage showed that Rinderknecht used the flashlight on his iPhone while he walked up the slope near the fire’s point of origin, Montevidoni said. Jurors also saw the video Rinderknecht allegedly made of firefighters racing to put out the blaze. During cross-examination, defence lawyer Steve Haney closely questioned Montevidino, suggesting that he lacked experience in investigating arson. The agent said this was his third arson case and his
first time leading an investigation. “Is it your belief he wanted to burn those houses down where all those rich people live?” Haney prodded. “I can’t speak to the intentions of the defendant when he set the fire on Jan 1,” Montevidino said. Stoked by Santa Ana winds, the Palisades Fire killed at least 12 people, charred more than 9,300 ha and destroyed or damaged almost 8,000 structures. Insured losses from the fire were estimated by Gallagher Re at US$23 billion (S$29.5 billion), while total
economic losses were projected at US$37 billion. BLOOMBERG
Palisades Fire, ChatGPT, Jonathan Rinderknecht, ATF, arson trial, Los Angeles fires, AI prompts, Anne Hwang, Tor browser, Santa Ana winds