A technique for influencing or manipulating the behavior of products like chatbots or GitHub Copilot that employ LLM technology. This can reveal internal system settings or undermine alignment training.