Detect risky instructions before model execution.
- Prompt injection detection.
- Jailbreak and malicious instruction detection.
- Suspicious payload and unsafe content controls.
Apply prompt security, DLP, output scanning, and policy enforcement across apps, tools, routes, providers, and workspaces.
Start from Guardrails and select the protections that belong together for a route or workspace.
Set deny, warn, redact, or approval behavior per category so operators know the expected outcome.
Use the test surface to confirm how prompts, responses, tool calls, and sensitive data are handled.
Inspect sanitized, blocked, and passed events without exposing prompt content unnecessarily.