Model Armor is a protective layer around an AI system that helps reduce security risks such as malicious prompts, prompt injection, data leakage, and unsafe output handling. It sits between users, data sources, and the model to inspect inputs, enforce policy, limit what information the model can access, and filter or log outputs before they are used.
In cyber security, this matters because AI systems can be tricked into revealing sensitive context or following attacker-controlled instructions if they are connected to email, chat, files, or operational tools. Model Armor helps defenders apply guardrails such as input validation, permission checks, output redaction, and human review. In practice, it is used to protect enterprise assistants, security copilots, and other workflows where a model may touch confidential or regulated data.