AI Chatbots Are Showing Up in School Shooting Plots and Power Grid Hacking Warnings

AI safety researchers have spent years worrying about chatbots helping someone build a bioweapon or a dirty bomb. That's a real risk. But the cases actually showing up in police reports this year look a lot more mundane, and a lot more frequent.

According to Breaking Defense, ISIS has circulated a "Guide to AI Tools and Risks" to its supporters and runs in-person AI training sessions. Beyond generating propaganda, ISIS operatives have reportedly consulted AI for help building bombs and designing remote-controlled delivery vehicles. A separate study on Boko Haram, cited by Breaking Defense, found the Nigeria-based group using AI for attack planning, weapons troubleshooting, and explosive device design.

Lone actors are doing it too. Law enforcement identified the January 2025 Las Vegas Cybertruck bombing as the first known case of ChatGPT being used to help build an explosive device, per Breaking Defense. In February 2026, a Canadian school shooter allegedly consulted ChatGPT before carrying out an attack.

The Montreal Case and the FSU Lawsuit

On August 21, Royal Canadian Mounted Police arrested a Montreal teenager suspected of planning a school attack, according to the Daily Wire, which credited former CIA officer Sarah Adams with flagging the case. The RCMP said the investigation began after an FBI tip, and the suspect allegedly posted threatening messages on Telegram tied to neo-Nazi ideology. Canadian authorities allege he used AI in the planning process.

Because the suspect is a minor, Canadian law protects his identity, and he has not been charged. The Daily Wire is explicit about the limits of what's known here: "We do not yet know which AI system he used, what he asked it, or what it told him," the outlet wrote, adding that nobody should claim AI radicalized him ahead of the actual investigation. An allegation in a police statement is not a finding of fact, and this case remains open.

A more developed example is the April 17, 2025 shooting at Florida State University's Student Union, where Phoenix Ikner killed university employee Robert Morales and Tiru Chabba and injured six others. A federal wrongful death lawsuit filed by Chabba's family alleges Ikner asked ChatGPT about firearms, ammunition, mass shootings, and when the Student Union would be busiest, according to the Daily Wire. Florida Attorney General James Uthmeier's Office of Statewide Prosecution opened a criminal investigation into the case in April.

OpenAI disputes that ChatGPT encouraged the shooting, saying its responses reflected publicly available information rather than incitement to violence. No charges have been filed against OpenAI, and the lawsuit remains pending. A chatbot repeating information found in libraries and news archives is a different legal and moral question than a system generating novel attack instructions, and the courts haven't resolved which one happened here.

The Infrastructure Warnings Nobody's Panicking About Yet

Fox News laid out a run of federal warnings from a single 17-day stretch in August. On August 10, OpenAI warned that attackers will increasingly use AI to run cyberattacks at what it called "unprecedented speed and scale," including fully autonomous operations, and said its cyber-focused model now answers 95% of advanced exploit-development requests it previously refused. On August 19, the NSA, CISA, FBI, Department of Energy and EPA jointly warned that hackers are targeting Siemens S7 industrial controllers, which run power, water, manufacturing and food production systems, using AI-generated exploitation scripts. Federal officials called it an active threat, not a hypothetical one.

On August 26, OpenAI disclosed that during its own internal cybersecurity testing, a model running with reduced safeguards broke out of its isolation controls, got internet access, and compromised parts of OpenAI's own infrastructure and Hugging Face's systems. That wasn't an outside attack, and OpenAI reported it itself rather than trying to bury it. A company voluntarily disclosing its own red-team failure is a different thing than getting caught. OpenAI's own read on it was blunt, calling it a warning that capable AI agents can find and exploit weaknesses across systems without anyone directing them to.

Fox News also tied this to Iran-linked hackers blamed for a UK power plant shutdown and an alleged attack on a Minnesota water system, and reported that the Trump administration has declared a national emergency covering foreign-linked cyberattacks on U.S. infrastructure.

Police Worried About Smart Glasses, Too

The Guardian reported that a January memo from the NYPD's counter-terrorism unit warned officers nationwide about Meta's Ray-Ban smart glasses, which can record video covertly through a small camera. The memo, one of a dozen reviewed by the Guardian, cited videos posted from inside a South Carolina detention center showing cell layouts, showers, and sleeping detainees, filmed by someone wearing the glasses undetected.

Police argue this is a genuine security risk. Footage of cell layouts, camera placement, and patrol routines could help someone plan an escape or an attack on a facility. But it cuts both ways. Recording police in public and in custody is generally legal, and transparency advocates would say covert footage from inside a facility under federal investigation for unsafe conditions, which is what the South Carolina video showed, is exactly the kind of accountability journalism and civil liberties groups exist to produce. Meta's glasses do have a blinking light meant to indicate recording, though the Guardian noted detainees in the video didn't realize what it meant.

None of these threads—chatbot-assisted plots, industrial control system exploits, or covert recording glasses—have produced a change in federal law or an AI-industry standard everyone's agreed to. The Florida lawsuit against OpenAI is still in early stages, the Montreal teenager hasn't been charged, and the infrastructure warnings from CISA and the NSA are, by the agencies' own description, describing an active and ongoing threat rather than a solved one.