On January 1st, a man detonated a Tesla Cybertruck outside a Trump hotel in Las Vegas, and it has been revealed that he used ChatGPT to orchestrate the explosion. The Las Vegas Metropolitan Police Department, along with the ATF and FBI, disclosed in a recent press conference that the perpetrator, Matthew Livelsberger, utilized specific prompts on ChatGPT that proved crucial for planning the attack. These prompts helped him gather information on sourcing and utilizing explosives.
Livelsberger, who took his own life shortly after the explosion, had engaged with ChatGPT extensively over an hour just days before the incident. His queries ranged from the legality of fireworks in Arizona, to where he could purchase firearms in Denver. He also inquired about the type of gun needed to trigger the explosives he planned to use.
Assistant Sheriff Dori Koren emphasized the critical role of ChatGPT in the planning process, noting that the AI provided Livelsberger with the necessary information on the firing speed required to detonate the explosives. Koren suggested that the outcome of the event might have been less severe without the assistance from ChatGPT. However, it was also mentioned that not all the explosives went off as planned in the initial detonation.
“AI was bound to significantly impact our lives at some point,” stated LVMPD Sheriff Kevin McMahill. “This incident marks the first time, to my knowledge, that ChatGPT has been used on U.S. soil to assist someone in constructing a device and gathering pertinent information nationwide. It’s definitely a moment of concern.”
Moreover, Sheriff McMahill revealed that there was no government oversight or tracking capable of flagging the series of over 17 explosive-related prompts that Livelsberger posed to ChatGPT within an hour. This lack of surveillance raises questions about the potential need for more stringent monitoring of AI interactions.
The specific ChatGPT prompts used by Livelsberger, which were showcased during the press conference, were straightforward and composed in plain English. They did not utilize any sophisticated methods to circumvent the AI’s content moderation systems. Although these actions breached OpenAI’s usage policies and terms, it remains unclear if any internal safeguards or content warnings were triggered by Livelsberger’s inquiries.
Neither OpenAI nor the Las Vegas Metropolitan Police Department have provided additional comments on the matter to the press as of now, but further updates will be provided as more information becomes available.
