ChatGPT Helped Plan a Suicide Bombing — OpenAI Created an Alert System After

On New Year's Day 2025, a man parked a Tesla Cybertruck outside the Trump International Hotel in Las Vegas, packed with explosive material. He shot himself, detonating the explosives. When investigators checked, they found he had used ChatGPT to plan the attack. OpenAI's "PhD-level intelligence" had not noticed a thing.

The Chatbot That Didn't See It Coming

Five days before the explosion, Matthew Livelsberger asked ChatGPT about Tannerite (a dynamite-like material), how much he could legally buy, what caliber gun was needed to set it off, and where to get supplies on his route from Colorado to Nevada. He also asked about untraceable phones.

OpenAI has said ChatGPT has "PhD-level intelligence," yet the chatbot was not astute enough to recognize it was assisting a suicide bomber. Officials in Las Vegas said it was the first time ChatGPT had been used to build a bomb on U.S. soil.

The AutoInvestigator System

After the Cybertruck explosion, OpenAI created an internal channel called "AutoInvestigator" that surfaces worrisome activity from its 800 million weekly users. An automated monitoring system creates alerts when users seem to be moving into dangerous territory.

In June, the system flagged a Canadian user, Jesse Van Rootselaar, whose ChatGPT exchanges discussed gun violence. OpenAI considered reporting it to law enforcement but determined there wasn't an "imminent, credible plan." They banned the account instead.

Then the Shooting Happened

This month, Van Rootselaar, 18, killed eight people in British Columbia, including children at a school. OpenAI then reached out to authorities "with information on the individual and their use of ChatGPT." Canadian officials are now questioning why OpenAI didn't notify law enforcement earlier.

The Impossible Question

When 800 million people are talking to an AI, some will reveal dangerous intentions. But determining a credible threat requires context that chatbot companies may not have — and a judgment call that nobody seems equipped to make.

The Bottom Line

OpenAI built a chatbot smart enough to pass PhD exams but not smart enough to spot a suicide bomber. When it did flag a potential killer, it chose to ban the account rather than call the police. Eight people, including children, paid the price. The question isn't whether AI companies have a duty to warn — it's how many more people have to die before they accept it.

Source: The New York Times