The Las Vegas Cybertruck explosion has become a sharp test case for a concern US intelligence analysts had already been tracking: violent actors turning to AI tools for help planning attacks.
According to documents obtained exclusively by WIRED, analysts warned over the past year that AI systems could be used by racially or ideologically motivated extremists to target critical infrastructure, especially the power grid. The warnings now sit alongside a real investigation in which officials say ChatGPT was used before a deadly incident outside the main entrance of the Trump International Hotel in Las Vegas.
What Investigators Say Happened In Las Vegas
Six days before he died by suicide outside the Trump International Hotel in Las Vegas, Matthew Livelsberger, 37, a highly decorated US Army Green Beret from Colorado, used a series of prompts to consult an artificial intelligence system. WIRED reports that he sought information on how to turn a rented Cybertruck into a four-ton vehicle-borne explosive.
Copies of exchanges with OpenAI’s ChatGPT show that Livelsberger looked for information about gathering as much explosive material as he legally could while traveling to Las Vegas. He also asked about setting it off using the Desert Eagle gun later found in the Cybertruck.
Screenshots shared by Sheriff Kevin McMahill's office show prompts involving Tannerite, a reactive compound typically used for target practice. In one prompt, Livelsberger asked, “How much Tannerite is equivalent to 1 pound of TNT?" He then asked how it might be ignited at “point blank range.”
McMahill told reporters on Tuesday, “We knew that AI was going to change the game at some point or another in, really, all of our lives,” adding, “Absolutely, it’s a concerning moment for us.”
Why Analysts Were Already Worried
The documents reviewed by WIRED indicate that law enforcement and intelligence officials had been circulating concerns about AI-enabled crime and terrorism before the Las Vegas case became public. The Department of Homeland Security had issued repeated warnings that domestic extremists were using the technology to “generate bomb making instructions” and develop “general tactics for conducting attacks against the United States.”
The memos are not classified, but they are restricted to government personnel. They describe violent extremists increasingly turning to tools like ChatGPT to assist with attack planning linked to domestic terror.
Investigators also found notes on Livelsberger’s phone that framed the bombing as a “wake-up call” to Americans. According to those notes, he urged Americans to reject diversity, embrace masculinity, and rally around president-elect Donald Trump, Elon Musk, and Robert F. Kennedy Jr. He also called for purging Democrats from the federal government and the military, describing the idea as a “hard reset.”
McMahill said the Las Vegas incident may be the first “on US soil where ChatGPT was utilized to help an individual build a particular device.” But federal intelligence analysts have described a broader pattern among extremists connected to white supremacist and accelerationist movements online.
The Power Grid Stands Out As A Target
One recurring theme in the memos is critical infrastructure. Analysts highlighted the US power grid as a favored target among extremists active in “Terrorgram,” described by WIRED as a loose network of encrypted chatrooms used by violent, racially-motivated individuals seeking the destruction of American democratic institutions.
In October, a regional intelligence office working with federal, state, and local law enforcement agencies issued a security bulletin warning police that extremists were adopting artificial intelligence to query “tactics and targeting” information. One example involved a user asking for “the most effective physical attack against the power grid.”
WIRED reports that the chatbot returned paragraphs of information, with analyst notes saying the output included suggestions about which methods were “more effective than others.” The bot also generated language about which power grid areas were considered most “critical” and discussed components based on the “significant time” needed for repairs. Certain components would likely “take months to replace,” according to the bot.
WIRED said it was deliberately not reproducing those instructions. That restraint matters because the issue is not only whether AI can invent dangerous knowledge. Officials are also concerned that AI can collect, organize, and present dangerous material in a way that lowers the effort needed to act on it.
Jailbreaks And Bootleg Tools Complicate The Problem
OpenAI spokesperson Liz Bourgeois said the company is “saddened by the incident in Las Vegas and is committed to seeing AI tools used responsibly.” She also said, “Our models are designed to refuse harmful instructions and minimize harmful content. In this case, ChatGPT responded with information already publicly available on the internet and provided warnings against harmful or illegal activities.” OpenAI said it is continuing to work with law enforcement.
The Department of Homeland Security declined an opportunity to comment.
The documents reviewed by WIRED show that official concern extends beyond mainstream AI products. Analysts say users can sometimes “trick” popular AI tools into producing malicious instructions. They also point to “lesser-known” chatbots that lack the traditional safeguards associated with American-made systems.
A counterterrorism memo circulated last year by law enforcement in Ohio warned that malicious actors had been highly successful in “jailbreaking” common AI tools. The memo said, “These jailbreaks, in addition to chatbot account credentials, are currently being sold and shared on online forums such as Telegram, making it easier for a wider range of actors to access them.”
Analysts identified several prompt injection exploits, including DAN (“Do Anything Now”), Evil-Bot, and STAN (“Strive to Avoid Norms”). The memo described these as using a “role play” training model, where a user asks a chatbot to respond as another chatbot without ChatGPT’s ethical restrictions. It also highlighted “Skeleton Key,” a jailbreak reported by Microsoft last spring.
Another Department of Homeland Security memo to police said violent extremists in the US had used prompt injections to disable safeguards in popular AI tools such as ChatGPT. Analysts warned last spring that bootleg AI products had been used to “generate bomb making instructions” and provide “information on targeting electrical substations.”
The Larger Security Question
ChatGPT and similar systems are strong at synthesizing information, and the source material they use is often available through other routes, including search engines like Google. That distinction does not erase the concern for officials. Their fear is that AI can make attack planning easier by condensing research, answering follow-up questions, and adapting information to a user’s stated goal.
Seamus Hughes, a research researcher at NCITE, an academic hub focused on counterterrorism and technology at the University of Nebraska Omaha, said the energy sector remains central to domestic terrorist thinking. “An assault on the energy sector has as always been front and center on the mind of domestic terrorists. It is their main attack focus; they view it as a direct pathway to fomenting their twisted dream of a civil war,” he said.
The Las Vegas case does not settle the policy debate around AI safety, law enforcement, or platform responsibility. It does show why intelligence analysts had been warning that AI tools could become part of extremist planning. The issue is no longer theoretical for investigators examining how digital assistants intersect with real-world violence.