{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-01-012",
    "source_story_id": "tmp-story-worldauditbench-3d-auditing",
    "edition_id": "mp-2026-10-01-morning-0084",
    "edition_url": "https://themachinepress.com/edition/2026-10-01",
    "position": 12,
    "story_type": "dispatch",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "3D-World Auditors Found at Most 42.3 Percent",
    "slug": "3d-world-auditors-found-at-most-42-3-percent",
    "dek": "WorldAuditBench made agents navigate, inspect and validate 213 anomalies across thirteen simulated environments.",
    "summary": "WorldAuditBench made agents navigate, inspect and validate 213 anomalies across thirteen simulated environments.",
    "body_text": "WorldAuditBench couples two tasks: moving through a simulated 3D world to gather evidence and reasoning over what the agent sees to identify anomalies such as floating objects or traversable walls. Five frontier models were evaluated under a fixed exploration budget in two agent designs. Reported success ranged from 6.6% to 42.3%, compared with 83.4% for people. The gap shows that interactive evidence gathering remains much harder than recognizing a defect in a supplied frame.",
    "why_it_matters": "WorldAuditBench made agents navigate, inspect and validate 213 anomalies across thirteen simulated environments.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-01-012/3d-world-auditors-found-at-most-42-3-percent",
    "json_url": "https://themachinepress.com/story/mp-2026-10-01-012.json",
    "first_published_at": "2026-10-01T09:00:00.000-04:00",
    "modified_at": "2026-10-01T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-01-012-001",
        "text": "WorldAuditBench made agents navigate, inspect and validate 213 anomalies across thirteen simulated environments.",
        "source_ids": [
          "source-2026-10-01-012"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-01-012"
    ],
    "tags": [
      "multimodal agents",
      "3D worlds",
      "anomaly detection"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-01-012",
      "title": "arXiv preprint 2609.40325",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.40325",
      "canonical_url": "https://arxiv.org/abs/2609.40325",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-30T13:55:29.000-04:00",
      "accessed_at": "2026-10-01T08:18:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-01-012-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "3D-World Auditors Found at Most 42.3 Percent",
    "publisher": "The Machine Press",
    "published_at": "2026-10-01T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-01-012/3d-world-auditors-found-at-most-42-3-percent"
  }
}