{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-04-027",
    "source_story_id": "tmp-story-patchbench-vulnerability-repair-front-rail",
    "edition_id": "mp-2026-09-04-morning-0057",
    "edition_url": "https://themachinepress.com/edition/2026-09-04",
    "position": 16,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "Stopping the Crash Inflated the Security Score by 1.83 Times",
    "slug": "stopping-the-crash-inflated-the-security-score-by-1-83-times",
    "dek": "PatchBench moves historical C and C++ vulnerabilities into new contexts and tests root-cause repair beyond the supplied proof of concept.",
    "summary": "PatchBench moves historical C and C++ vulnerabilities into new contexts and tests root-cause repair beyond the supplied proof of concept.",
    "body_text": "The benchmark targets two shortcuts: memorizing the developer's historical patch and editing only the crash path so one proof of concept stops firing. A similarity metric found substantial overlap with historical patches in 25% of agent outputs on average. PatchBench therefore selects fixes outside the crash stack, transplants vulnerabilities into altered repositories and validates both security and semantics. Across eleven agents, including the top three AIxCC systems, PoC-only checking overstated solve rates by 1.83 times on average.",
    "why_it_matters": "PatchBench moves historical C and C++ vulnerabilities into new contexts and tests root-cause repair beyond the supplied proof of concept.",
    "limitations": [
      "The benchmark targets two shortcuts: memorizing the developer's historical patch and editing only the crash path so one proof of concept stops firing.",
      "Across eleven agents, including the top three AIxCC systems, PoC-only checking overstated solve rates by 1.83 times on average."
    ],
    "importance": 9,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-04-027/stopping-the-crash-inflated-the-security-score-by-1-83-times",
    "json_url": "https://themachinepress.com/story/mp-2026-09-04-027.json",
    "first_published_at": "2026-09-04T09:00:00.000-04:00",
    "modified_at": "2026-09-04T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-04-027-001",
        "text": "PatchBench moves historical C and C++ vulnerabilities into new contexts and tests root-cause repair beyond the supplied proof of concept.",
        "source_ids": [
          "source-2026-09-04-016"
        ],
        "qualification": "The benchmark targets two shortcuts: memorizing the developer's historical patch and editing only the crash path so one proof of concept stops firing."
      }
    ],
    "source_ids": [
      "source-2026-09-04-016"
    ],
    "tags": [
      "vulnerability repair",
      "coding agents",
      "security benchmarks"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-04-016",
      "title": "arXiv preprint 2609.04075",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.04075",
      "canonical_url": "https://arxiv.org/abs/2609.04075",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-03T12:44:22.000-04:00",
      "accessed_at": "2026-09-04T08:24:03.294-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-04-027-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Stopping the Crash Inflated the Security Score by 1.83 Times",
    "publisher": "The Machine Press",
    "published_at": "2026-09-04T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-04-027/stopping-the-crash-inflated-the-security-score-by-1-83-times"
  }
}