{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-02-005",
    "source_story_id": "tmp-story-kalibench-cybersecurity-cli",
    "edition_id": "mp-2026-10-02-morning-0085",
    "edition_url": "https://themachinepress.com/edition/2026-10-02",
    "position": 5,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "No Open Model Cleared 42 Percent on Unhinted Security Commands",
    "slug": "no-open-model-cleared-42-percent-on-unhinted-security-commands",
    "dek": "KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.",
    "summary": "KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.",
    "body_text": "KaliBench measures whether models can translate analyst intent into executable command lines, with deterministic canonicalization, alias-aware scoring, sandbox execution and human refinement. Across 24 configurations, no open-weight model exceeded 42 percent exact-command accuracy without tool hints. Fine-tuning and verifiable-reward training substantially improved an eight-billion-parameter model to performance comparable with a tested 685-billion-parameter mixture-of-experts model. Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised.",
    "why_it_matters": "KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.",
    "limitations": [
      "Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-005/no-open-model-cleared-42-percent-on-unhinted-security-commands",
    "json_url": "https://themachinepress.com/story/mp-2026-10-02-005.json",
    "first_published_at": "2026-10-02T09:00:00.000-04:00",
    "modified_at": "2026-10-02T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-02-005-001",
        "text": "KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.",
        "source_ids": [
          "source-2026-10-02-005"
        ],
        "qualification": "Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised."
      }
    ],
    "source_ids": [
      "source-2026-10-02-005"
    ],
    "tags": [
      "cybersecurity",
      "tool use",
      "benchmarks"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-02-005",
      "title": "arXiv preprint 2610.02206",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.02206",
      "canonical_url": "https://arxiv.org/abs/2610.02206",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-01T13:59:55.000-04:00",
      "accessed_at": "2026-10-02T08:44:06.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-02-005-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "No Open Model Cleared 42 Percent on Unhinted Security Commands",
    "publisher": "The Machine Press",
    "published_at": "2026-10-02T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-005/no-open-model-cleared-42-percent-on-unhinted-security-commands"
  }
}