Guardrails and content policies

Guardrails are structural constraints on LLM inputs and outputs that enforce safety, correctness, and policy compliance. They operate at multiple layers: the API gateway, the prompt construction layer, the model itself, and post processing pipelines.