Hacked Hardware Can Force AI Chatbots to Give Dangerous Answers

Researchers found that vulnerabilities in the physical components of large-scale AI systems can allow attackers to manipulate chatbot outputs to provide incorrect or harmful information.