Skip to main content

Review Queue

The Review Queue surfaces AI interactions that your validators have flagged as requiring a human decision before the system can act on them. Rather than automatically blocking or allowing a borderline interaction, the enforcement policy routes it here for a reviewer to approve, reject, or escalate.

Queue Overview​

The page header shows the current Pending count — the number of items awaiting a human decision. Items are presented in the Pending tab by default. The All tab shows the complete history including already-reviewed items.

Each item in the queue shows:

  • Risk Score — A numeric score from 0–100 reflecting how many validators flagged the interaction and how severe those flags were.
  • REVIEW REQUIRED badge — Indicates the item is still awaiting a decision, with the timestamp it entered the queue.
  • Trigger Tags — Labels showing which validator categories fired on this interaction (e.g., pii, secrets, compliance, banned_topics, response_completeness).

Reviewing an Item​

Expanding a queue item reveals the full context needed to make a decision:

  • Input Message — The complete prompt or input sent to the AI agent.
  • Validator Breakdown — A per-validator result list showing every validator that ran against this interaction, its pass/fail status, and the specific reason it passed or failed. This breakdown covers all validator types: security, privacy, reliability, and compliance validators.

After reviewing the context, three actions are available:

  • Approve — The interaction is marked compliant and logged. The originating workflow may proceed.
  • Reject — The interaction is blocked and logged as a violation.
  • Escalate — The item is escalated to a higher-level reviewer or incident response workflow.

Note: All three decisions are recorded in the Audit Log with the reviewer's action and timestamp.

How Items Enter the Queue​

Items appear in the Review Queue when a Policy's enforcement action for a borderline result is set to flag_and_escalate. This is configured per policy — see Policies for details on enforcement action configuration.

Best Practices​

  • Assign clear ownership for the Review Queue in your security operations process — pending items affect your compliance rate on the Compliance Dashboard.
  • Use the trigger tags to triage quickly: items tagged pii or secrets are typically higher priority than response_completeness.
  • If the same validator is generating a high volume of false positives, refine its configuration in Validators rather than approving items manually at scale.
  • For items that reveal a pattern (e.g., a specific agent consistently triggering banned_topics), address the root cause in the agent's policy or guardrail configuration.