Guardrails
Guardrails are security policies that inspect LLM requests and responses to detect and block harmful, policy-violating, or inappropriate content before it reaches the model or the user. You can apply prompt guards to the request phase, the response phase, or both.
To learn more about guardrails, see the following topic.
To set up guardrails, check out the following guides.
Use custom regex patterns and built-in PII detectors to filter LLM requests and responses.
Use the OpenAI Moderation API to detect harmful content across categories including hate, harassment, and violence.
Apply AWS Bedrock Guardrails to filter LLM requests and responses for policy-violating content.
Apply Google Cloud Model Armor templates to sanitize LLM requests and responses.
Integrate your own content safety logic by forwarding requests and responses to a custom webhook.
Run prompt guards in sequence, creating defense-in-depth protection.
To track guardrails and content safety, see the following guide.
