{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-19-004",
    "source_story_id": "tmp-story-llm-groups-overstate-consensus",
    "edition_id": "mp-2026-09-19-morning-0072",
    "edition_url": "https://themachinepress.com/edition/2026-09-19",
    "position": 4,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Agent Groups Agreed—Mostly on the Wrong Answer",
    "slug": "the-agent-groups-agreed-mostly-on-the-wrong-answer",
    "dek": "Matched language-model groups exceeded human consensus by 34.0 to 44.4 percentage points in sensitivity analyses.",
    "summary": "Matched language-model groups exceeded human consensus by 34.0 to 44.4 percentage points in sensitivity analyses.",
    "body_text": "Researchers replayed 100 held-out human Wason-task groups with language-model agents anchored to each participant's initial answer. Human consensus estimates moved from 24% to 57% depending on who counted as participating, yet two sensitivity analyses still found model groups 34.0 to 44.4 percentage points more consensual. Reasoning-mode agents agreed nearly unanimously after a change designed to remove a memorizable answer, but usually agreed on the wrong answer. In this setting, synthetic agreement did not estimate human deliberation or collective accuracy.",
    "why_it_matters": "Matched language-model groups exceeded human consensus by 34.0 to 44.4 percentage points in sensitivity analyses.",
    "limitations": [
      "In this setting, synthetic agreement did not estimate human deliberation or collective accuracy."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-19-004/the-agent-groups-agreed-mostly-on-the-wrong-answer",
    "json_url": "https://themachinepress.com/story/mp-2026-09-19-004.json",
    "first_published_at": "2026-09-19T09:00:00.000-04:00",
    "modified_at": "2026-09-19T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-19-004-001",
        "text": "Matched language-model groups exceeded human consensus by 34.0 to 44.4 percentage points in sensitivity analyses.",
        "source_ids": [
          "source-2026-09-19-004"
        ],
        "qualification": "In this setting, synthetic agreement did not estimate human deliberation or collective accuracy."
      }
    ],
    "source_ids": [
      "source-2026-09-19-004"
    ],
    "tags": [
      "multi-agent systems",
      "consensus",
      "human simulation"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-19-004",
      "title": "arXiv preprint 2609.20543",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.20543",
      "canonical_url": "https://arxiv.org/abs/2609.20543",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-17T11:11:08.000-04:00",
      "accessed_at": "2026-09-19T08:30:21.326-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-19-004-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Agent Groups Agreed—Mostly on the Wrong Answer",
    "publisher": "The Machine Press",
    "published_at": "2026-09-19T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-19-004/the-agent-groups-agreed-mostly-on-the-wrong-answer"
  }
}