AI agent security: 10 checks before shipping

AI agent security starts with verified behavior. Use 10 checks for prompt injection, permissions, sandboxing, regression tests, and shutdown controls.