Skip to main content
Anthale evaluates the guardrails configured in a policy and returns one runtime action your application must enforce.

Evaluation flow

  1. Anthale receives content, direction, and metadata.
  2. Anthale runs the guardrails configured for that policy and request path.
  3. Each configured guardrail evaluates the content against the policy settings.
  4. Anthale resolves the result into one returned action.

Action priority

When Anthale resolves multiple hits, block wins over redact, redact wins over detect, and detect wins over allow.

Configured guardrail actions

At the policy level, guardrails are configured to detect, redact, or block when they trigger. The exact choices depend on the guardrail.

Returned runtime actions

allow

Continue the flow unchanged.

detect

Continue the flow and log the event for review.

redact

Continue with Anthale’s sanitized content instead of the raw content.

block

Stop the request or response and return a safe fallback.

Failure handling

Your service still decides whether transport failures fail open or fail closed. Anthale returns the runtime result shape, but your application owns resilience policy. Use one shared enforcement layer or middleware, keep fallback responses standardized, and log every returned action with request metadata. Avoid app-specific overrides that weaken stronger actions.

Common mistakes

The common failures are sending the original content after a redact, treating detect as if nothing happened operationally, and handling the same action differently across services.

Next steps

Next, read API Reference Overview or the Python Direct SDK Path.
Last modified on March 19, 2026