Breaking LLMs From the Outside In: A Field Guide to Prompt Injection and Jailbreaking

Large language models don't have a firewall. They have an instruction hierarchy, a soft, learned sense of "the system prompt outranks the user, and the user outranks whatever text shows up inside a do