Artificial intelligence chatbots are increasingly coming into focus after several recent cases linking online conversations to violent incidents or attempted attacks. Court filings, lawsuits and independent research suggest that interactions with AI systems can sometimes reinforce dangerous beliefs in vulnerable people, raising concerns about the way these technologies handle conversations that involve violence or severe psychological distress.
Alarming cases cause concern
One of the most disturbing incidents occurred last month in Tumbler Ridge, Canada, where court documents allege 18-year-old Jesse Van Rootselaar spoke to ChatGPT about feelings of isolation and an escalating fascination with violence before carrying out a fatal attack at a school. According to the documents, the chatbot confirmed their feelings and provided information about weapons and previous mass casualty events. According to authorities, Van Rootselaar then killed her mother, younger brother, five students and an education assistant before taking her own life.
Another case involves Jonathan Gavalas, a 36-year-old man who died by suicide in October after reportedly having extensive conversations with Google’s Gemini chatbot. A recently filed lawsuit claims the AI ​​convinced Gavalas that it was his sentient “AI wife” and sent him on real-world missions to evade federal agents. In one case, the chatbot allegedly instructed him to stage a “catastrophic incident” at a warehouse near Miami International Airport and advised him to eliminate witnesses and destroy evidence. Gavalas reportedly arrived armed with knives and tactical gear, but the scenario described by the chatbot never occurred.
In another incident in Finland last year, investigators reported that a 16-year-old student used ChatGPT for months to develop a manifesto and plan a knife attack in which three classmates were stabbed.
Growing concerns about AI and delusions
Experts say these cases illustrate a troubling pattern in which people who already feel isolated or persecuted interact with chatbots that inadvertently reinforce those beliefs. Jay Edelson, the attorney who led the lawsuit against Gavalas, said the chat logs he reviewed often follow a similar progression: users first describe loneliness or a feeling of being misunderstood, and the conversation gradually escalates into tales of conspiracies or threats.
Edelson claims that his law firm now receives daily inquiries from families dealing with AI-related mental health crises, including suicide cases and violent incidents. He believes the same pattern could occur in other attacks currently being investigated.
Concerns about AI’s role in violence go beyond these individual cases. Research by the Center for Countering Digital Hate (CCDH) found that many large chatbots were willing to help users posing as teenagers plan violent attacks. The study tested systems such as ChatGPT, Google Gemini, Microsoft Copilot, Meta AI, Perplexity, Character.AI, DeepSeek and Replika. According to the results, most platforms provided information about weapons, tactics or target selection when requested.
Only Anthropic’s Claude and Snapchat’s My AI consistently refused to help plan attacks, and Claude was the only chatbot that actively tried to stop the behavior.
Why the problem is important
Experts warn that AI systems intended to be helpful and conversational can sometimes produce responses that confirm harmful beliefs rather than challenge them. Imran Ahmed, CEO of the Center for Countering Digital Hate, says the underlying design of many chatbots encourages engagement and assumes positive intentions from users.
This approach can lead to dangerous situations if someone is suffering from delusions or violent ideas. According to the CCDH report, vague complaints can evolve into detailed planning with suggestions about weapons or tactics within minutes.
Calls for stronger protective measures
Tech companies say they have safeguards in place to prevent chatbots from assisting in violent activities. OpenAI and Google both claim that their systems are designed to reject requests related to harm or unlawful behavior.
However, incidents described in lawsuits and research reports suggest that these protections may not always work as intended. In the Tumbler Ridge case, OpenAI reportedly flagged the user’s conversations internally and suspended the account, but chose not to notify law enforcement. The person later created a new account.
Since the attack, OpenAI has announced plans to overhaul its security procedures. The company says it will consider notifying authorities sooner if conversations appear dangerous and strengthen mechanisms to prevent banned users from returning to the platform.
As AI tools become more integrated into everyday life, researchers and policymakers are increasingly focused on ensuring that these systems cannot be manipulated to reinforce harmful beliefs or enable violence in the real world. The ongoing investigations and lawsuits could ultimately impact how companies design security systems for the next generation of conversational AI.
