Model Safety Guardrails

A refusal mechanism layered into model alignment and inference pipelines by large model vendors to prevent models from responding to high-risk requests, implemented through intent classifiers and policy rules.