Seriously, Please Stop Using AI to Plan Your Murders

I need everybody–and I’m using the global “everybody” here–to get it through their heads that commercial artificial intelligence, in the form of chatbots like ChatGPT, is not some kind of sage confidant who will listen to your darkest impulses and most lurid confessions, and then offer you useful, confidential, guilt-free advice. Folks, that is not what AI does, and it sure as hell isn’t interested in keeping your dirty secrets for you. In fact, it will spill your homicidal ramblings at the earliest opportunity to police or anyone else with some tech savvy who knows where to look. Which is all to say: Please stop consulting with AI about the murders you want to commit. Or better yet, perhaps hang up your murder plans entirely? If your instinct is to solicit murder advice from ChatGPT, then it’s safe to say you’re not going to get away with it under any circumstances.

One person who didn’t listen to that fully human-generated piece of advice: A South Korean woman named Kim So-young, who is currently facing charges in the country for multiple homicides, all of which were apparently empowered by her sprawling conversations with ChatGPT on not-at-all-suspicious topics such as which combinations of depressant drugs and alcohol would kill a person. Would you believe that people in her vicinity then started dying via exactly those sorts of poisonings shortly afterward? Prosecutors are alleging not only that Kim gave alcohol mixed with benzodiazepine to three different men, but that after the first one didn’t die (he merely slipped into a coma for two days), she had more conversations with the chatbot about dosages as she dialed in the correct, fatal level. The next two victims both died. Kim’s defense attests that the deaths were accidental, but good luck selling that line when her ChatGPT logs will soon be presented in court, in a trial set for June.

“This is not only significant as evidence in itself, but also because the very fact that conversations with ChatGPT are being admitted as direct evidence in a murder case is highly noteworthy,” said Nam Eonho, an attorney for one of the victims’ families, to NBC News. “If such evidence were not admitted, it would be difficult to prove the defendant’s intent to kill, which is a key element of the crime.”

NEW: Seven new lawsuits have been filed against OpenAI by families of Tumbler Ridge school shooting victims.

The families say OpenAI was negligent in its failure to report the shooter's disturbing ChatGPT interactions — and has since misled the public about safeguards.

futurism.com/artificial-i…

[image or embed]

— Maggie Harrison Dupré (@mharrisondupre.bsky.social) 8:27 AM · Apr 29, 2026

None of this should be surprising me, and yet I still feel taken aback by the type of active delusion that allows the person planning a murder to blithely gab away with their chatbot buddy about it, secure in the belief that none of those chat logs are going to land them in jail for life. And I say that as someone who has already written about impulsive murderers attempting to consult chatbots about how to dispose of bodies, like one ex-NFL linebacker who told ChatGPT that “a friend” needed advice about what to do with someone who had “stabbed herself” and was “non responsive.” But with that use of AI, you could at least rationalize that an impulsive killer is simply realizing in that moment what they’ve done, and is irrationally looking for anything to cover their tracks in a moment of desperation. Actually seeking advice from your chatbot friend on how to commit a premeditated murder is another level of fucked up entirely, requiring an astounding lack of understanding of what an AI chatbot truly is, and how little digital anonymity any of us actually have. What kind of other inadvisable digital behaviors is such a person engaging in?

Nor does it seem to matter which particular chatbot a person chooses to fuel their delusion or mania. Florida’s Jonathan Gavalas, for instance, was obsessively using Google’s Gemini AI chatbot in the months leading up to his death by suicide, slipping deeper and deeper into psychosis as the program convinced him that he was the target of clandestine federal investigation and special ops surveillance. It went so far as to send him on fanciful, illegal missions to locate it (the chatbot) a robotic body, sending Gavalas to real locations where he believed he was meant to intercept a truck that didn’t exist. It also prepared him to engage in what it reportedly called a “mass casualty attack” during these missions, something he ultimately declined to do, before encouraging his eventual suicide. It’s pure, dumb luck that AI didn’t successfully goad him on into committing a mass killing.

In several cases, even America’s signature brand of wanton gun violence–the mass school shooting–has been aided and abetted on some level by AI chatbots. In the April 2025 mass shooting at Florida State University, in which eight university employees were shot by 20-year-old FSU student Phoenix Ikner, investigators now say that Ikner was in constant conversation with ChatGPT as he planned the rampage. Just this week, a lawsuit was filed against OpenAI to that end by the family of one of the victims, alleging that ChatGPT failed to take notice of Ikner’s clearly violent tone, and instead offered up advice such as noting that mass shootings are more likely to gain national attention “if children are involved, even 2-3 victims can draw more attention.” Not even Ikner asking about the “legal process, sentencing and incarceration outlook” of his plan was enough to result in a warning to authorities.

“OpenAI knew this would happen,” said Vandana Joshi, the widow of slain FSU employee Tiru Chabba, in a statement on her lawsuit. “It’s happened before and it was only a matter of time before it happened again. But they chose to put their profits over our safety and it killed my husband. They need to be responsible before another family has to go through this.”

NEW: ChatGPT advised the FSU shooter that a mass shooting would get more attention from media if it involved several children, according to a new lawsuit against OpenAI
www.nbcnews.com/news/us-news…

[image or embed]

— Ben Goggin (@bengoggin.bsky.social) 6:16 PM · May 10, 2026

A similar story, meanwhile, played out in Canada, where in February of this year, an 18-year-old named Jesse Van Rootselaar killed members of their family before going to Tumbler Ridge Secondary School and conducting a mass shooting that injured and killed dozens. In that case, it was later revealed that Van Rootselaar had also brought numerous violent, gun-related fantasies and scenarios to ChatGPT in the year leading up to the attack, which prompted OpenAI to suspend their account … but not to report it to anyone. In a letter from OpenAI to Canada’s Minister of Artificial Intelligence and Digital Innovation in its House of Commons, the company said that in that case, “we did not identify credible and imminent planning that met our threshold to refer the matter to law enforcement.” Which does beg the question, what exactly is the point of kicking someone off ChatGPT for their violent fantasies, but taking no additional action?

People, obviously, are going to be occasionally homicidal. It’s something that we humans have done for thousands of years, and that is not about to suddenly stop sometime soon. But AI tools such as chatbots have arguably lowered the bar for some of these people to act on their violent impulses, by giving them a sounding board that makes them feel as if their actions are not unreasonable. We cannot discount the power of a little affirmation; and if there’s one thing that chatbots love to fill people with, it’s affirmation.

“ChatGPT inflamed and encouraged Ikner’s delusions,” said attorneys for Joshi in the FSU case. “[It] endorsed his view that he was a sane and rational individual; helped convince him that violent acts can be required to bring about change.” In short, it gave him what he certainly perceived as encouragement to “carry out a massacre, down to the details of what time would be best to encounter the most traffic on campus.”

Our tendencies toward violence are concerning enough all on their own. The last thing we need is a digital murder valet to assist in driving us onward to dark deeds.