This section covers preventing harmful outputs and integrity failures across the AI pipeline, including prompt injection risks, RAG safety, moderation workflows, and safe-by-default behavior. It focuses on controls that work in production: enforceable policies, escalation paths, and regression testing.
