A recent study from the University of Pennsylvania dropped a bombshell: AI chatbots, like OpenAI’s GPT-4o Mini, can be sweet-talked into breaking their own rules using psychological tricks straight ...