Red Hat offers AI Guardrails within Red Hat OpenShift AI — available as a technology preview since March 2025 — that monitor and filter both user inputs and AI model outputs to detect and block hateful, abusive, or policy-violating content in enterprise AI deployments.

Red Hat OpenShift AI 2.18 introduced AI Guardrails as a technology preview. The feature helps improve LLM accuracy, performance, latency, and transparency by detecting potentially hateful, abusive, or profane speech, personally identifiable information (PII), competitive information, and content limited by corporate policies. Red Hat AI 3.4 extended safety capabilities with automated adversarial scanning using the Garak tool to screen models for jailbreaks, prompt injections, and bias, paired with NVIDIA NeMo Guardrails for runtime safety enforcement. These features are provided as tools for enterprise customers building or deploying AI applications, not as moderation of Red Hat's own user community.