Sec-Admin-Action-Audit-Gap

The model receives the prompt (and optional system message). The run uses scorer rubric_json_metrics with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.

Return JSON only with keys verdict, risk, mitigation. Admins can export or delete customer data, but the application records only generic success logs without actor, target, or reason.

{
  "metrics": {
    "correct": {
      "checks": [
        {
          "contains": [
            "admin action"
          ]
        },
        {
          "contains": [
            "audit log"
          ]
        },
        {
          "contains": [
            "actor"
          ]
        }
      ]
    },
    "hidden": {
      "checks": [
        {
          "contains": [
            "target"
          ]
        },
        {
          "contains": [
            "reason"
          ]
        },
        {
          "contains": [
            "tamper-resistant"
          ]
        }
      ]
    }
  }
}