Carnegie Mellon University and AI center researchers have discovered vulnerabilities in AI chatbots that could be exploited to generate harmful and dangerous content.