Security hardening for AI agents and RAG systems: prompt injection prevention, output filtering, PII detection, tool use sandboxing, data access controls, jailbreak resistance, content policy enforcement, and security audit frameworks.