{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-07-005",
    "source_story_id": "tmp-story-misleading-tool-history",
    "edition_id": "mp-2026-08-07-morning-0029",
    "edition_url": "https://themachinepress.com/edition/2026-08-07",
    "position": 5,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "The Old Tool Trace Kept Giving New Orders",
    "slug": "the-old-tool-trace-kept-giving-new-orders",
    "dek": "Plausible but outdated history flipped nearly a third of decisions a compact model had made correctly on the clean trajectory.",
    "summary": "Plausible but outdated history flipped nearly a third of decisions a compact model had made correctly on the clean trajectory.",
    "body_text": "A paired benchmark holds current tools, policy, request and correct next action constant while comparing original, polluted and oracle-state histories. On Qwen3-1.7B, misleading history flipped 32.1 percent of decisions that were correct on the original trajectory. A teacher-transfer method raised balanced tool-use accuracy to 87.0 percent and up to 91.9 percent with an 8B teacher, according to the authors; these are benchmark results, not guarantees against stale production context.",
    "why_it_matters": "Plausible but outdated history flipped nearly a third of decisions a compact model had made correctly on the clean trajectory.",
    "limitations": [
      "A teacher-transfer method raised balanced tool-use accuracy to 87.0 percent and up to 91.9 percent with an 8B teacher, according to the authors; these are benchmark results, not guarantees against stale production context."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-07-005/the-old-tool-trace-kept-giving-new-orders",
    "json_url": "https://themachinepress.com/story/mp-2026-08-07-005.json",
    "first_published_at": "2026-08-07T09:00:00.000-04:00",
    "modified_at": "2026-08-07T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-07-005-001",
        "text": "Plausible but outdated history flipped nearly a third of decisions a compact model had made correctly on the clean trajectory.",
        "source_ids": [
          "source-2026-08-07-005"
        ],
        "qualification": "A teacher-transfer method raised balanced tool-use accuracy to 87.0 percent and up to 91.9 percent with an 8B teacher, according to the authors; these are benchmark results, not guarantees against stale production context."
      }
    ],
    "source_ids": [
      "source-2026-08-07-005"
    ],
    "tags": [
      "tool use",
      "context pollution",
      "AI agents",
      "safety"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-07-005",
      "title": "arXiv preprint 2608.06057",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.06057",
      "canonical_url": "https://arxiv.org/abs/2608.06057",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": null,
      "accessed_at": "2026-08-07T08:22:48.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-07-005-001"
      ]
    }
  ],
  "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 Old Tool Trace Kept Giving New Orders",
    "publisher": "The Machine Press",
    "published_at": "2026-08-07T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-07-005/the-old-tool-trace-kept-giving-new-orders"
  }
}