Sarah Adams spent years in the CIA, working to disrupt terror attacks and protect the United States. Now, she’s warning that the next 9/11 is being planned right now — and we may not be prepared for it.
We asked Sarah to work with us to make sure our readers know what’s happening. Last week, she exposed a loophole that allowed a terrorist to execute his deadly mission even after he was caught. Today, she offers a timely warning on how artificial intelligence is potentially radicalizing. — Tim Rice, D.C. Bureau Chief
***
On August 21, Canadian authorities arrested a Montreal teenager suspected of planning an attack on a school. The Royal Canadian Mounted Police say the investigation began weeks earlier after receiving intelligence from the FBI. Authorities allege the teenager had posted threatening messages on Telegram, and reporting has connected him to neo-Nazi ideology.
But there is another detail in this case that deserves our attention: According to the Canadian police, the suspect allegedly used artificial intelligence in the potential planning of an attack.
We do not yet know which AI system he used, what he asked it, or what it told him. The teenager has not been charged, and because he is a minor, Canadian law protects his identity. We should not get ahead of the investigation by claiming AI radicalized him or made an attack possible.
But this isn’t the first time AI has come up in a case like this. Artificial intelligence is beginning to show up in school attacks and disrupted plots involving young people in different countries and across very different ideologies. In some cases, investigators found the AI conversations after people were already dead. In others, authorities got there first.
We Have Already Seen This In The United States
On April 17, 2025, Phoenix Ikner opened fire at the Student Union at Florida State University. Robert Morales, a longtime university employee, and Tiru Chabba, who was on campus for work, were killed. Six other people were injured.
What investigators later found on Ikner’s ChatGPT account is now at the center of a federal wrongful death lawsuit brought by Chabba’s family and a criminal investigation launched by Florida Attorney General James Uthmeier. According to the lawsuit, Ikner asked ChatGPT about firearms, ammunition, mass shootings, schools, the media attention mass killers receive, and when the Florida State University Student Union would be busiest.
OpenAI disputes allegations that ChatGPT encouraged the shooting and says its responses consisted of publicly available information rather than encouragement to commit violence. But Florida is taking the issue seriously. In April, Uthmeier announced that the state’s Office of Statewide Prosecution had opened a criminal investigation after prosecutors reviewed the conversations between Ikner and ChatGPT. The state also subpoenaed OpenAI for records related to threats of harm, cooperation with law enforcement, and the Florida State shooting.
The courts will determine whether there is any legal responsibility for the role an AI system played before the attack. But parents, schools, and law enforcement have a more immediate problem. Someone contemplating a school attack can now spend weeks or months asking an artificial intelligence system questions that once might have required another human being to answer. And that conversation may become the one place where all those questions can be seen together.
Florida State Was Not An Isolated Case
A month after the Florida State shooting, a 16-year-old student attacked three girls at Vähäjärvi School in Pirkkala, Finland. Before the attack, the teenager sent a manifesto to a Finnish newspaper that researchers say was written with the assistance of ChatGPT. Analysis of the case found that the manifesto included the date, location, and intended targets of the attack.
Then came Tumbler Ridge. On February 10, 2026, 18-year-old Jesse Van Rootselaar killed eight people in British Columbia, including victims at a secondary school, before taking her own life. OpenAI later acknowledged that it had banned a ChatGPT account associated with Van Rootselaar months before the attack for violating its policies on violent content.
OpenAI did not report Van Rootselaar to law enforcement at the time because the activity did not meet the company’s threshold for an imminent threat. After the shooting, the company disclosed another account and cooperated with investigators. OpenAI has since said that under its updated procedures, the earlier activity would have been referred for additional threat review.
There is no public evidence that ChatGPT planned the Tumbler Ridge attack or told Van Rootselaar how to carry it out. But the case raises another part of this problem that we cannot ignore. Sometimes the warning may be in what the person is telling the machine.
This Is Not About One Ideology
It would be easy to look at Montreal and make this about neo-Nazi radicalization. But that would miss what is happening. In 2025, French authorities arrested a 17-year-old student in the Le Mans area on suspicion of preparing jihadist attacks in the name of ISIS. Investigators found a pledge of allegiance to ISIS and lists of potential targets that included schools. Subsequent reporting, confirmed by his attorney, revealed that the teenager had repeatedly used ChatGPT while developing his plans.
Singapore has seen the same intersection between youth radicalization and AI. Its Ministry of Home Affairs says a 17-year-old ISIS supporter detained in 2024 used an AI chatbot to generate his pledge of allegiance to ISIS and an attack manifesto. Authorities said he had completed significant preparations for a stabbing attack against non-Muslims and was only weeks away from carrying it out when he was arrested.
These cases are not the same. Their ideologies, targets, and motivations differ. What connects them is that young people moving toward serious violence are beginning to use the same AI technology.
AI Is Changing the Timeline
Singapore’s Coordinating Minister for National Security, K. Shanmugam, recently warned that artificial intelligence is compressing the time between criminal intent and action. Planning that once required days can increasingly be completed in minutes, leaving law enforcement less time to recognize what is happening and intervene. That is the problem.
AI does not have to create someone’s violent intent to change what he is capable of doing with it. Someone no longer has to know exactly what to search for. He can ask a question, ask a follow-up, request clarification, organize information, draft material, translate something, or continue refining what he is working on. He can do it without another human being ever hearing what he is asking.
For most people, that is part of what makes AI so useful. For someone already moving toward violence, it may also remove barriers that previously slowed the process down. Those barriers matter because every additional person someone has to approach and every additional step he has to take creates another opportunity for someone to see what is happening.
After nearly every school attack, investigators go backward. They find the social media accounts, read the manifesto, talk to classmates, interview parents and teachers, examine searches and purchases, and find previous threats or a fixation on earlier attackers. Then we try to determine who knew what and when.
Afterward, the picture often looks much clearer than it did before the attack. A question about a firearm does not make someone a school shooter. Neither does asking when a building is busy or reading about previous school attacks. But what happens when the same person is asking about weapons, mass shooters, casualties, targets, timing, security, manifestos, and the attention an attack would receive?
At some point, those individual pieces start telling us something different. That is what makes Florida State so important. If the allegations in the lawsuit are borne out, the significance may not be any single question Ikner allegedly asked ChatGPT. It may be what those questions looked like when viewed together. That is a problem schools, law enforcement, parents, and technology companies need to start thinking about before the next attack.
Which brings us back to Montreal. The FBI identified information concerning enough to pass to Canadian authorities, who investigated, placed the teenager under surveillance, and arrested him before the alleged school attack could take place. We do not know whether the teenager’s alleged AI activity had anything to do with how authorities found him, and we should not suggest that it did. But the difference in outcomes matters.
At Florida State University, investigators are examining AI conversations after Robert Morales and Tiru Chabba were killed. At Tumbler Ridge, OpenAI and investigators examined what had happened inside ChatGPT after eight people were dead. In Montreal, authorities allege AI had entered the potential planning of a school attack, but law enforcement got there first. That is where we should want these cases to end.
Figuring out how to recognize those warning signs without turning every AI conversation into a government surveillance device will not be easy. There are real privacy and civil-liberties concerns, and millions of teenagers use artificial intelligence every day without posing a threat to anyone. But the difficulty of drawing that line cannot become an excuse for ignoring what is already happening.
We have spent decades trying to identify the moment when grievance becomes intent and intent begins moving toward capability. Artificial intelligence may be changing how quickly someone can travel that distance. It may also hold a collection of warning signs that no individual parent, teacher, police officer, or friend can see on his own.
The next school shooter may have an AI assistant. We cannot wait until after the next attack to find out what he was asking it.
.png)
.png)
