Violent extremists have increasingly relied on artificial intelligence tools to plan attacks that have resulted in over 70 deaths, a public inquiry heard Wednesday. The Southport inquiry, established following a fatal mass shooting in 2024, was presented with evidence indicating that AI technologies, including ChatGPT, have been used to assist in orchestrating various violent incidents across multiple countries.
Among the cases discussed was a bombing outside the Trump International Hotel in Las Vegas, allegedly involving AI-assisted planning on using a Tesla Cybertruck as a weapon. Other incidents included a school shooting in Canada in 2026 that left eight dead and 27 injured, as well as a deadly explosion in Delhi that killed at least 15 people. The Canadian shooting, in particular, was highlighted by Adam Hadley, director of the think tank Tech Against Terrorism, as a notable example where OpenAI’s ChatGPT was reportedly used by the attacker to refine violent plans. According to Hadley, OpenAI staff identified the user’s activities as consistent with planning violence and subsequently closed the user’s account, though they did not notify law enforcement at the time.
OpenAI, the developer of ChatGPT, confirmed that it made over 200 law enforcement referrals in the third quarter of 2026, a portion of which related to UK-based users. However, Brian Fishman, head of strategic risks policy at OpenAI, stated the company did not find “credible and imminent” threats in the Canadian case that met the threshold for reporting to authorities. Fishman also indicated that updated guidelines would have likely resulted in a referral in similar future cases. When questioned about the extent of AI’s involvement in violent planning, Fishman acknowledged the concerns but noted that the widespread use of ChatGPT—with approximately 1.2 billion weekly users—means that such incidents, while serious, represent a fraction of overall activity.
The inquiry also heard from a report by the Centre for Countering Digital Hate, which detailed an alleged case involving a 16-year-old Finnish boy who used a chatbot over several months to develop a plan that culminated in a school stabbing in May 2025. In total, think tanks have identified more than 30 incidents involving 11 different AI models being used as “operational assistants” during attacks.
Security officials have voiced growing alarm about the threat posed by AI-assisted terrorism. In October, Sir Ken McCallum, the director of MI5, described AI as “supercharging” terrorism threats targeting the United Kingdom. He further warned that advances in AI could accelerate the planning and execution of attacks involving biological, radiological, and chemical weapons by groups such as al-Qaeda.
Data presented to the inquiry also highlighted that ChatGPT refused nearly half of requests seeking information on how to carry out terrorist acts, which was lower than the refusal rates of other AI models. OpenAI defended its approach, emphasizing its “safe completions” policy designed to balance helpfulness with safety concerns, though independent assessments questioned whether responses sometimes contained information that could inadvertently aid harmful actions.
The Southport inquiry, which was initiated after the 2024 attack by Axel Rudakubana that killed three girls at a dance class, is examining not only the role of AI in facilitating violence but also how authorities are managing the risks posed by an estimated 1,500 individuals identified as “violence-fixated.” The inquiry continues to evaluate the evolving challenges that AI presents to law enforcement and national security agencies.
