{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-09-001",
    "source_story_id": "tmp-lead-white-box-deception-probes",
    "edition_id": "mp-2026-10-09-morning-0091",
    "edition_url": "https://themachinepress.com/edition/2026-10-09",
    "position": 1,
    "story_type": "lead",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "Deception Left a Signal the Transcript Didn’t Show",
    "slug": "deception-left-a-signal-the-transcript-didn-t-show",
    "dek": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
    "summary": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
    "body_text": "A convincing transcript can leave a monitor without the evidence it needs. Researchers trained white-box probes on a new deception dataset and aggregated signals across model layers and tokens. On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught. They also tested deception whose ground truth required careful elicitation or knowledge of training data. In one hidden-goal evaluation, the probes distinguished transcripts containing the model’s true goal from alternatives with up to 99.7% AUC. The released FIBS dataset supports further testing. These are reported evaluations, including open-weight models, rather than a universal honesty detector or proof that a monitor can read intentions.",
    "why_it_matters": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
    "limitations": [
      "On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught."
    ],
    "importance": 10,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-001/deception-left-a-signal-the-transcript-didn-t-show",
    "json_url": "https://themachinepress.com/story/mp-2026-10-09-001.json",
    "first_published_at": "2026-10-09T09:00:00.000-04:00",
    "modified_at": "2026-10-10T03:43:44.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-09-001-001",
        "text": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
        "source_ids": [
          "source-2026-10-09-001"
        ],
        "qualification": "On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught."
      }
    ],
    "source_ids": [
      "source-2026-10-09-001"
    ],
    "tags": [
      "deception detection",
      "white-box probes",
      "agent monitoring"
    ],
    "image_url": "https://themachinepress.com/issues/2026-10-09/lead-deception-probes.png",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-09-001",
      "title": "arXiv preprint 2610.12445, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12445v1",
      "canonical_url": "https://arxiv.org/abs/2610.12445v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:13.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-001-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Deception Left a Signal the Transcript Didn’t Show",
    "publisher": "The Machine Press",
    "published_at": "2026-10-09T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-001/deception-left-a-signal-the-transcript-didn-t-show"
  }
}