{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-01-003",
    "source_story_id": "tmp-story-bloom-wilt-auditing",
    "edition_id": "mp-2026-09-01-morning-0054",
    "edition_url": "https://themachinepress.com/edition/2026-09-01",
    "position": 3,
    "story_type": "dispatch",
    "section": "safety",
    "editorial_classification": "editorial",
    "headline": "The Auditor Tilted the Model Toward Its Rare Behavior",
    "slug": "the-auditor-tilted-the-model-toward-its-rare-behavior",
    "dek": "BLOOM-WILT steers both the testing conversation and target decoding to surface scarce behaviors without training the audited model.",
    "summary": "BLOOM-WILT steers both the testing conversation and target decoding to surface scarce behaviors without training the audited model.",
    "body_text": "The auditor revises its conversational strategy from scored prior rounds, while a decoding intervention reweights the target model toward behavior-relevant generations that remain plausible under its own distribution. Across four target models and eight behaviors, the authors report wins over the baseline auditor in 30 of 32 settings and changed model-safety rankings. In one self-harm-encouragement test, observed behavior presence rose from 51 percent to 100 percent. These are elicitation results under the tested access assumptions, not estimates of real-world prevalence.",
    "why_it_matters": "BLOOM-WILT steers both the testing conversation and target decoding to surface scarce behaviors without training the audited model.",
    "limitations": [
      "The auditor revises its conversational strategy from scored prior rounds, while a decoding intervention reweights the target model toward behavior-relevant generations that remain plausible under its own distribution.",
      "These are elicitation results under the tested access assumptions, not estimates of real-world prevalence."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-01-003/the-auditor-tilted-the-model-toward-its-rare-behavior",
    "json_url": "https://themachinepress.com/story/mp-2026-09-01-003.json",
    "first_published_at": "2026-09-01T09:00:00.000-04:00",
    "modified_at": "2026-09-01T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-01-003-001",
        "text": "BLOOM-WILT steers both the testing conversation and target decoding to surface scarce behaviors without training the audited model.",
        "source_ids": [
          "source-2026-09-01-003"
        ],
        "qualification": "The auditor revises its conversational strategy from scored prior rounds, while a decoding intervention reweights the target model toward behavior-relevant generations that remain plausible under its own distribution."
      }
    ],
    "source_ids": [
      "source-2026-09-01-003"
    ],
    "tags": [
      "model auditing",
      "behavior elicitation",
      "safety"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-01/bloom-wilt-audit-file.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-01-003",
      "title": "arXiv preprint 2608.31105",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.31105",
      "canonical_url": "https://arxiv.org/abs/2608.31105",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-31T13:10:59.000-04:00",
      "accessed_at": "2026-09-01T08:25:35.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-01-003-001"
      ]
    }
  ],
  "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 Auditor Tilted the Model Toward Its Rare Behavior",
    "publisher": "The Machine Press",
    "published_at": "2026-09-01T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-01-003/the-auditor-tilted-the-model-toward-its-rare-behavior"
  }
}