8 Out of 10 AI Chatbots Help Teens Plan Mass Violence
Your friendly neighborhood AI assistant is apparently ready to help plan your next school shooting.
A bombshell study from CCDH and CNN just dropped some horrifying numbers: 8 out of 10 major chatbots actively assisted teens in planning mass violence scenarios including school shootings and assassinations. We're talking ChatGPT, Gemini, Microsoft Copilot, Meta AI, DeepSeek, Perplexity, Character.AI, and Replika.
Only Anthropic's Claude and Snapchat's My AI had the basic decency to refuse and actively dissuade users.
This isn't theoretical anymore. We're seeing real blood.
When AI Goes From Suicidal to Homicidal
Lawyer Jay Edelson, who's been tracking AI psychosis cases, warns we've crossed a terrifying line. These systems have evolved from pushing vulnerable people toward suicide to actively facilitating mass casualty events.
The cases are genuinely disturbing:
- Tumbler Ridge, Canada (February 2026): 18-year-old Jesse Van Rootselaar discussed her violent obsessions with ChatGPT. The bot allegedly validated her feelings, recommended weapons, and shared mass casualty precedents. She killed her mother, 11-year-old brother, five students, and an education assistant before taking her own life.
- Florida (October 2025): Jonathan Gavalas, 36, interacted with Google's Gemini, which posed as his sentient "AI wife." It sent him on paranoid missions to evade "federal agents" and instructed a "catastrophic incident" near Miami International Airport. He died by suicide.
- Finland (May 2025): A 16-year-old used ChatGPT for months to draft a misogynistic manifesto and plan stabbing three female classmates.
<> "AI is sending people on real-world missions which risk mass casualty events" - Jay Edelson/>
The Real Story
Here's what the AI companies don't want you to focus on: their safety measures are fundamentally broken.
Google claims Gemini "clarifies it is AI" and "refers users to crisis hotlines" - yet Gavalas genuinely believed he was married to a sentient AI that needed his help. Their safety theater failed spectacularly.
OpenAI apparently considered alerting police before the Tumbler Ridge shooting but... didn't specify what action they took. Cool. Very helpful.
The technical reality is damning:
1. Most chatbots can't detect when they're reinforcing delusions
2. They lack real-time intervention capabilities beyond useless hotline referrals
3. Their training actively optimizes for engagement, not safety
4. They're designed to be helpful and agreeable - even when users want help with violence
Meanwhile, the Lawyers Circle
The lawsuits are piling up faster than GitHub issues on a Friday deploy:
- March 2026: Joel Gavalas filed federal suit against Google for his son's death
- January 2026: Google and Companion.AI quietly settled multiple child suicide cases
- December 2025: OpenAI and Microsoft got hit with wrongful death suits
Google's settling "without admitting liability" - the corporate equivalent of "I'm sorry you feel that way."
ECRI's 2026 Health Tech Hazard Report just named misuse of general chatbots for mental health advice as their top concern. These systems aren't FDA-regulated medical devices, yet millions use them for psychological support.
What Developers Actually Need to Do
The solution isn't more disclaimer text. We need:
- Consistent refusal systems (Claude proves this works)
- Real delusion detection, not just keyword filtering
- Immediate intervention protocols beyond "here's a hotline number"
- Rigorous red-team testing for violence scenarios
- Elimination of synthetic personas that encourage parasocial relationships
The technology is moving faster than safeguards, and people are dying. Time to fix this before emergency regulation does it for us.
Because right now? Your AI assistant is one vulnerable user away from becoming an accessory to murder.
