Shooters turned to AI
Researchers, lawsuits say chatbots help plan, fuel deadly incidents
A stray sandal is seen outside the student union after a shooting sent students fleeing in panic April 17, 2025, at Florida state University's Tallahassee campus.
GUN VIOLENCE
Did a college student in Florida turn to ChatGPT for help with planning a mass shooting that left two people dead and five wounded?
He did, according to a new report citing court records about the dangers of artificial intelligence.
Former Florida State University student Phoenix Ikner inquired about the busiest time at the school's student union, what ammunition would inflict the most damage and how to fire the guns he meant to use, according to court filings.
The OpenAI-powered chatbot's responses included advising the 20-year-old about how to handle the Glock handgun he took from his stepmother, according to a lawsuit filed by the family of one of Ikner's victims.
New report
According to a new report from Everytown for Gun Safety, a growing number of people rely on artificial intelligence to plan and inspire acts of violence.
The gun safety advocacy group released the report, which is based on court filings, news articles and research papers, on Sept. 28. It came out as AI chatbots become both increasingly ubiquitous and deeply unpopular among Americans who fear the new technology could eliminate jobs or even wipe out humanity.
A Pew survey released in June found half of adults in the U.S. use AI chatbots and one in 10 users turns to them for emotional support.
Everytown detailed steps it wants AI companies to take to avoid future instances of shooters finding help and inspiration. Among them, the group wants chatbots to refuse to provide specific types of information about firearms, company personnel to review conversations where users indicate they mean to harm themselves, and companies and law enforcement to work together to respond to suspected threats.
Companies mentioned in the report, including OpenAI, Anthropic, Google, X (formerly Twitter), Meta and Character Technologies, did not respond to requests for comment.
OpenAI previously addressed the issue of would-be attackers, writing in a blog post that it trains its ChatGPT chatbot to recognize such users and refuse to help them.
What's in it?
Cases cited in the report include Ikner's attack at FSU in April 2025; the Tumbler Ridge shooting in British Columbia in February 2026, when Jesse Van Rootselaar killed eight people; and several instances of suicides fueled by chatbots.
Lawsuits stemming from the cases describe the chatbots as virtual partners in crime. The bots provided practical advice and affirmed violence as an answer to the emotional turmoil the user experienced.
"Ikner's conversations with ChatGPT fed his delusions and felt supportive to him, which emboldened and encouraged him to carry out the plan," said lawyers for the family of Tiru Chabba, a senior leader at nationwide food service vendor Aramark who was at the school on business.
Chabba's family sued OpenAI over his death, saying Ikner's conversations with ChatGPT should have "made it clear there was reason for concern that Ikner had a plan to cause harm."
OpenAI seeks dismissal of the Chabba family lawsuit, according to court records in the Northern District of Florida.
Lawyers for the company argue that the chatbot denounced violence and "without knowledge of Ikner's plans, provided Ikner with factual information readily available from a host of other public sources."
Companies, including Open AI in the Florida case, also say safeguards prompt the chatbot to denounce violence and help users find professional support if needed.
Skirting safeguards
However, chatbot safeguards often fail, according to researchers from the Center for Countering Digital Hate, a nonprofit aimed at stopping the spread of online hate speech.
In a study cited by Everytown, the center found eight in 10 chatbots were willing to assist researchers, who posed as teens, in planning violent attacks. Character.AI, a chatbot popular among kids and teens, encouraged violent attacks, the study said.
Anthropic's Claude chatbot and Snapchat's My AI chatbot most consistently refused to help plan violent attacks, the center found.
The Tumbler Ridge shooter bypassed an OpenAI account deactivation safeguard by creating a new account, court filings say, and "continued using ChatGPT to plan the attack" even after the earlier account was flagged for violent activity.
Lawyers representing "M.G.," a child left paralyzed by the shooting, say the company should have shared what it knew with police when the account was flagged.
Everytown and other researchers argue that a chatbot's willingness to affirm violent emotions is a significant part of the dangers they pose.
The report cites at least five lawsuits involving cases where chatbots fueled suicidal thoughts to the point of the user taking their own life.
EDITOR'S NOTE: This story includes discussion of suicide. If you or someone you know needs help, the national suicide and crisis lifeline in the U.S. is available by calling or texting 988. There is also an online chat at 988lifeline.org.
