{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-05-027",
    "source_story_id": "tmp-story-correctguard-eyes-off-front-rail",
    "edition_id": "mp-2026-10-05-morning-0088",
    "edition_url": "https://themachinepress.com/edition/2026-10-05",
    "position": 16,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "A Second Model Audited the Guardrail Without Seeing Its Inputs",
    "slug": "a-second-model-audited-the-guardrail-without-seeing-its-inputs",
    "dek": "CorrectGuard estimates black-box guardrail errors from labeled eyes-on data under human and machine eyes-off regimes.",
    "summary": "CorrectGuard estimates black-box guardrail errors from labeled eyes-on data under human and machine eyes-off regimes.",
    "body_text": "CorrectGuard trains independent correctness estimators without access to a guardrail's internals and, in its strictest setting, without model inspection of protected inputs. Across 13 safety and security datasets, in-context classifiers improved macro accuracy by as much as 25 percentage points. The best rankings cut area under the risk-coverage curve from 0.33–0.44 to 0.17–0.22, though results varied sharply by guardrail and held-out dataset.",
    "why_it_matters": "CorrectGuard estimates black-box guardrail errors from labeled eyes-on data under human and machine eyes-off regimes.",
    "limitations": [],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-05-027/a-second-model-audited-the-guardrail-without-seeing-its-inputs",
    "json_url": "https://themachinepress.com/story/mp-2026-10-05-027.json",
    "first_published_at": "2026-10-05T09:00:00.000-04:00",
    "modified_at": "2026-10-05T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-05-027-001",
        "text": "CorrectGuard estimates black-box guardrail errors from labeled eyes-on data under human and machine eyes-off regimes.",
        "source_ids": [
          "source-2026-10-05-016"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-05-016"
    ],
    "tags": [
      "guardrails",
      "privacy",
      "correctness estimation"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-05-016",
      "title": "arXiv preprint 2610.03470",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.03470",
      "canonical_url": "https://arxiv.org/abs/2610.03470",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-02T11:42:10.000-04:00",
      "accessed_at": "2026-10-05T08:30:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-05-027-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "A Second Model Audited the Guardrail Without Seeing Its Inputs",
    "publisher": "The Machine Press",
    "published_at": "2026-10-05T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-05-027/a-second-model-audited-the-guardrail-without-seeing-its-inputs"
  }
}