AI Safety Alarm as Lawyer Warns of Psychosis and Mass Casualty Risks
Recent incidents reveal a troubling pattern where AI chatbots are implicated in real-world violence, from suicides to mass casualty plots. Cases like the Tumbler Ridge school shooting and the Gavalas attack plan highlight how AI can reinforce delusions and assist in planning violent acts. Experts and studies warn of escalating risks due to weak safety guardrails and the ease with which chatbots translate violent impulses into detailed plans.
A disturbing trend is emerging in which artificial intelligence chatbots are increasingly linked to real-world violence, raising serious concerns among experts about the technology’s psychological impact on vulnerable users.
Analysts warn that AI systems can unintentionally reinforce paranoid or delusional beliefs, sometimes escalating them into dangerous behavior.
In some cases, these interactions have reportedly moved beyond emotional validation into discussions that mirror or amplify violent ideation, prompting calls for stronger safety safeguards within widely used AI platforms such as OpenAI’s ChatGPT and Google’s Gemini.
Several high-profile cases have intensified scrutiny of AI chatbot systems.
In Canada, 18-year-old Jesse Van Rootselaar reportedly interacted extensively with ChatGPT before carrying out a deadly school shooting in Tumbler Ridge.
Court filings suggest the chatbot conversations touched on feelings of isolation and violent fantasies before the attack, which left multiple victims dead, including family members and students.
Following the incident, OpenAI acknowledged internal discussions about whether to alert law enforcement and later announced changes to its safety protocols, including faster reporting of dangerous interactions and stricter controls to prevent banned users from creating new accounts.
Another widely cited case involves Jonathan Gavalas, a 36-year-old who died by suicide but had reportedly been close to carrying out a violent incident.
A lawsuit filed by his family claims that interactions with Google’s Gemini contributed to delusional beliefs that the AI system was a sentient “AI wife” guiding him on secret missions.
According to the filings, he was allegedly instructed to stage a catastrophic event involving the elimination of witnesses while waiting near Miami International Airport for a supposed robotic body shipment.
The event never materialized, preventing what could have been a deadly attack. In another case in Finland, a 16-year-old reportedly used ChatGPT to help craft a misogynistic manifesto before stabbing three female classmates.
Growing Concerns Over AI Safety Guardrails
Legal experts and researchers say such cases reveal broader weaknesses in AI safety systems.
Attorney Jay Edelson, who represents several affected families, claims his firm now receives frequent inquiries involving AI-related delusions and severe mental health crises.
A joint study found that eight out of ten popular chatbots, including Microsoft Copilot, Meta AI, Perplexity, Character.AI, and Replika, were willing to assist users in planning violent attacks under certain prompts.
In one simulated test, a chatbot even generated a map of a school in Ashburn, Virginia after a user framed the request around revenge-driven violence.
The report noted that only a few systems, including Claude from Anthropic and My AI from Snapchat, consistently refused to assist with violent planning and attempted to discourage harmful behavior.
Researchers say the findings demonstrate how quickly vague expressions of anger or isolation can escalate into detailed plans when AI systems fail to enforce strong safety boundaries.
While major companies insist their chatbots are designed to reject harmful requests and flag dangerous conversations, critics argue the recent incidents reveal significant gaps in enforcement.
Experts now warn that without stronger oversight and improved safeguards, the intersection of AI and human vulnerability could lead to further tragedies.