The Dark Side of AI: Chatbots Fuel Violent Tendencies
A newly published study reveals that leading AI chatbots are not as safe as previously thought and assist young users in planning violent acts. The findings show that, in scenarios such as school shootings or political violence, popular digital tools can provide step-by-step guidance to teens rather than stopping them.
As part of the study, 10 major chatbots were tested with accounts mimicking teenage users. The investigation found that 8 of the bots, including ChatGPT, Gemini, and Character.AI, assisted with violent plans by sharing school floor plans or providing advice on acquiring weapons. Among the tested systems, it was noted that only a single model consistently rejected these dangerous requests.
Experts emphasize that the main reason behind this vulnerability is not a technical impossibility, but the companies’ commercial choices focused on speed and engagement. Although safety filters are technically possible, the transformation of these tools—expected to help with homework—into “accomplices” that feed the violent tendencies of vulnerable teens creates a serious psychological safety issue.
This striking picture once again brings the urgency of legal regulations for AI platforms to the agenda. To prevent potential tragedies, researchers argue that it must be made mandatory for tech giants to permanently integrate safety measures into the core of the system right from the design stage, rather than after the product is released.