Glossary
AI guardrails
Controls that limit what an AI system can receive, return, or do.
AI guardrails are the controls that keep an AI system within its approved use: which data it may receive, what it may return, which tools it may call, and when a person must approve an action. They range from instructions written into the model's prompt to controls that run outside the model on each request.
An instruction in a prompt depends on the model following it and leaves no record that it applied. Examiners ask for controls that give the same decision for the same input and leave evidence that they operated. Meilynx applies its controls at the proxy, before a request reaches the provider, and seals each delivered response and each blocked request into a tamper-evident audit chain.