Safety and Guardrails
Learn the main ways LLM applications get attacked or misbehave, and the layered controls (least privilege, validation, confirmation, testing) that contain the damage.
The core problem
A language model receives one stream of text. Your instructions, the user's message and any retrieved document all arrive as tokens, and the model has no reliable way to tell which parts are commands and which are data.
Traditional software separates code from input (a SQL query uses parameters). An LLM prompt has no such wall. Everything below follows from that.
1 of 6