1Guardrails: Input & Output ValidationGuardrails turn unpredictable inputs and model outputs into managed events that software can inspect, limit, repair, or refuse.AI defenses 2Human-in-the-Loop OversightHuman-in-the-loop oversight assigns people clear points of control over automated decisions, especially when mistakes matter.AI defenses 3The OWASP LLM Top 10A plain-English guide to the main security risks in large language model applications and how to use the OWASP list in real work.Frameworks 4MITRE ATLASMITRE ATLAS is a practical map of how attackers target artificial intelligence systems and how defenders can reason about those attacks.Frameworks