Guardrails overview

Guardrails inspect prompts and model output for injection, toxicity, PII, and related risks before a turn completes.

Surface	Role
Studio Governance → LLM Governance	Create profiles, Control library, attach to flow/agent
Workspace Governance → Guardrails	Analytics (Blocked / Logged, Decisions / Triggers)

Pro+ · workspace.governance.*. There is no separate Studio Guardrails rail in this IA.

Phinite Guardrails

Built-in Control library — Before / After the model, Session data.

Phinite, AWS Bedrock, Azure Content Safety, GCP Model Armor.

Entire flow or one agent.

End-to-end create → attach → verify.

Provider	What it is
Phinite	Built-in guard library (prompt injection, toxicity, PII, and more)
AWS Bedrock	Bedrock Guardrails — topics, PII, word filters, content policies
Azure Content Safety	Hate, self-harm, sexual, violence detection
GCP Model Armor	Prompt/response sanitization via Model Armor template