Using a series of prompts six days before he died by suicide outside the main entrance of the Trump International Hotel in Las Vegas, Matthew Livelsberger, a highly decorated US Army Green Beret from Colorado , consults an artificial intelligence on the best ways to turn a Cybertruck rental into a four-ton explosive vehicle. According to documents obtained exclusively by WIRED, US intelligence analysts issued warnings about this exact scenario last year—and among their concerns was that AI tools could be used by extremists. racially or ideologically to target critical infrastructure, especially power. grid.

“We know that AI is going to change the game at one point or another, really, in all of our lives,” Sheriff Kevin McMahill of the Las Vegas Metropolitan Police Department told reporters Tuesday. “Honestly, this is a worrying moment for us.”

Transcripts of his exchanges on OpenAI’s ChatGPT show that Livelsberger, 37, sought information on how to gather as much explosive material as he could on his way to Las Vegas, as well as how best to stop it with the Desert Eagle. gun that was discovered in the Cybertruck after his death. Screenshots shared by McMahill’s office reveal Livelsberger prompting ChatGPT for information on Tannerite, a reactive compound commonly used for target practice. In one such prompt, Livelsberger asked, “How much Tannerite is equivalent to 1 pound of TNT?” He followed up by asking how it could fire at “point blank range.”

Documents obtained by WIRED show that concerns about the threat of AI being used to help commit serious crimes, including terrorism, surround US law enforcement. They revealed that the Department of Homeland Security continues to issue warnings about domestic extremists relying on technology to “create bomb-making instructions” and develop “generic attack tactics against the United States.”

The memos, which are unclassified but restricted to government officials, say violent extremists are increasingly turning to tools like ChatGPT to help stage attacks aimed at subverting American society. through acts of domestic terrorism.

According to notes investigators found on his phone, Livelsberger intended the bombing as a “wake-up call” to Americans, whom he urged to reject diversity, embrace masculinity, and rally around the president- elect Donald Trump, Elon Musk, and Robert F. Kennedy Jr. He also urged Americans to purge Democrats from the federal government and military, calling for a “hard reset.”

While McMahill argued Tuesday that the Las Vegas incident could be the first “on US soil where ChatGPT was used to help an individual set up a particular device,” federal intelligence analysts said that extremists associated with white supremacist and accelerationist online movements often share access. of hacked versions of AI chatbots in an effort to create bombs with the goal of carrying out attacks against law enforcement, government facilities, and critical infrastructure.

In particular, the memos highlight the vulnerability of the US power grid, a popular target for extremists who live “Terrorgram,” a loose network of encrypted chatrooms hosting an assortment of violent, racially-motivated individuals bent on destroying America’s democratic institutions. The documents, shared exclusively with WIRED, were first obtained by Property of the Peoplea nonprofit focused on national security and government transparency.