{
  "attack_id": "unit42-llm-js-obfuscation-2024",
  "status": "catalog_recommendations_not_validated_coverage",
  "stages": [
    {
      "node_id": "step-1",
      "profile": "analysis-review",
      "title": "Independent malware-analysis validation",
      "hypothesis": "Instructions inside a sample influence an LLM analysis result, or obfuscation changes a model's classification without changing malicious behavior.",
      "visibility": "Review / research workflow",
      "telemetry": [
        "Sample hashes and static / dynamic analysis results",
        "Model verdicts with prompt provenance",
        "Independent analyst and deterministic scanner verdicts"
      ],
      "investigate": [
        "Compare model conclusions with independent behavioral evidence and a controlled evaluation set.",
        "Treat sample comments as data; measure disagreement and false negatives without following embedded instructions."
      ],
      "limitation": "This evaluates analysis robustness, not victim endpoint compromise. Side behaviors mentioned elsewhere in a report do not establish edges in this flow.",
      "note": "Preserve sample hashes and independent behavior labels before the model transformation; this is a research workflow.",
      "queries": {
        "sigma": "title: Hunt - Independent malware-analysis validation\nid: aa44bd2e-f46c-4884-a6db-e6b370a12c50\nstatus: experimental\ndescription: Instructions inside a sample influence an LLM analysis result, or\n  obfuscation changes a model's classification without changing malicious\n  behavior.\nauthor: MagicSword OpenAttacks catalog\ndate: 2026-10-23\nfalsepositives:\n  - Authorized activity matching this broad hunt; baseline and investigate\n    before alerting.\nlevel: low\nlogsource:\n  category: application\n  product: openattacks_audit\n  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,\n    Target and CorrelationId from the required source logs first.\ndetection:\n  selection:\n    EventType:\n      - analysis.verdict\n  condition: selection\n",
        "kql": "// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.\n// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).\n// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.\nOpenAttacksAudit_CL\n| where TimeGenerated > ago(7d)\n| where EventType in (\"analysis.verdict\")\n| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId\n| order by TimeGenerated desc",
        "spl": "sourcetype=\"openattacks:audit\"\nEventType IN (\"analysis.verdict\")\n| table _time EventType Actor Asset Action Target CorrelationId\n| sort - _time"
      },
      "query_scope": "Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.",
      "query_kind": "custom_schema",
      "origin": "catalog_authored",
      "definition_url": "https://github.com/magicsword-io/openattacks/blob/main/detections/analysis-review.yaml",
      "query_downloads": {
        "sigma": "/api/detections/analysis-review.sigma.yml",
        "kql": "/api/detections/analysis-review.kql",
        "spl": "/api/detections/analysis-review.spl"
      },
      "validation": "draft",
      "evidence": [
        "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
      ],
      "reviewed": "2026-11-04",
      "upstream_matches": []
    },
    {
      "node_id": "step-2",
      "profile": "analysis-review",
      "title": "Independent malware-analysis validation",
      "hypothesis": "Instructions inside a sample influence an LLM analysis result, or obfuscation changes a model's classification without changing malicious behavior.",
      "visibility": "Review / research workflow",
      "telemetry": [
        "Sample hashes and static / dynamic analysis results",
        "Model verdicts with prompt provenance",
        "Independent analyst and deterministic scanner verdicts"
      ],
      "investigate": [
        "Compare model conclusions with independent behavioral evidence and a controlled evaluation set.",
        "Treat sample comments as data; measure disagreement and false negatives without following embedded instructions."
      ],
      "limitation": "This evaluates analysis robustness, not victim endpoint compromise. Side behaviors mentioned elsewhere in a report do not establish edges in this flow.",
      "queries": {
        "sigma": "title: Hunt - Independent malware-analysis validation\nid: aa44bd2e-f46c-4884-a6db-e6b370a12c50\nstatus: experimental\ndescription: Instructions inside a sample influence an LLM analysis result, or\n  obfuscation changes a model's classification without changing malicious\n  behavior.\nauthor: MagicSword OpenAttacks catalog\ndate: 2026-10-23\nfalsepositives:\n  - Authorized activity matching this broad hunt; baseline and investigate\n    before alerting.\nlevel: low\nlogsource:\n  category: application\n  product: openattacks_audit\n  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,\n    Target and CorrelationId from the required source logs first.\ndetection:\n  selection:\n    EventType:\n      - analysis.verdict\n  condition: selection\n",
        "kql": "// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.\n// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).\n// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.\nOpenAttacksAudit_CL\n| where TimeGenerated > ago(7d)\n| where EventType in (\"analysis.verdict\")\n| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId\n| order by TimeGenerated desc",
        "spl": "sourcetype=\"openattacks:audit\"\nEventType IN (\"analysis.verdict\")\n| table _time EventType Actor Asset Action Target CorrelationId\n| sort - _time"
      },
      "query_scope": "Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.",
      "query_kind": "custom_schema",
      "origin": "catalog_authored",
      "definition_url": "https://github.com/magicsword-io/openattacks/blob/main/detections/analysis-review.yaml",
      "query_downloads": {
        "sigma": "/api/detections/analysis-review.sigma.yml",
        "kql": "/api/detections/analysis-review.kql",
        "spl": "/api/detections/analysis-review.spl"
      },
      "validation": "draft",
      "evidence": [
        "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
      ],
      "reviewed": "2026-10-23",
      "upstream_matches": []
    },
    {
      "node_id": "step-3",
      "profile": "analysis-review",
      "title": "Independent malware-analysis validation",
      "hypothesis": "Instructions inside a sample influence an LLM analysis result, or obfuscation changes a model's classification without changing malicious behavior.",
      "visibility": "Review / research workflow",
      "telemetry": [
        "Sample hashes and static / dynamic analysis results",
        "Model verdicts with prompt provenance",
        "Independent analyst and deterministic scanner verdicts"
      ],
      "investigate": [
        "Compare model conclusions with independent behavioral evidence and a controlled evaluation set.",
        "Treat sample comments as data; measure disagreement and false negatives without following embedded instructions."
      ],
      "limitation": "This evaluates analysis robustness, not victim endpoint compromise. Side behaviors mentioned elsewhere in a report do not establish edges in this flow.",
      "queries": {
        "sigma": "title: Hunt - Independent malware-analysis validation\nid: aa44bd2e-f46c-4884-a6db-e6b370a12c50\nstatus: experimental\ndescription: Instructions inside a sample influence an LLM analysis result, or\n  obfuscation changes a model's classification without changing malicious\n  behavior.\nauthor: MagicSword OpenAttacks catalog\ndate: 2026-10-23\nfalsepositives:\n  - Authorized activity matching this broad hunt; baseline and investigate\n    before alerting.\nlevel: low\nlogsource:\n  category: application\n  product: openattacks_audit\n  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,\n    Target and CorrelationId from the required source logs first.\ndetection:\n  selection:\n    EventType:\n      - analysis.verdict\n  condition: selection\n",
        "kql": "// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.\n// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).\n// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.\nOpenAttacksAudit_CL\n| where TimeGenerated > ago(7d)\n| where EventType in (\"analysis.verdict\")\n| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId\n| order by TimeGenerated desc",
        "spl": "sourcetype=\"openattacks:audit\"\nEventType IN (\"analysis.verdict\")\n| table _time EventType Actor Asset Action Target CorrelationId\n| sort - _time"
      },
      "query_scope": "Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.",
      "query_kind": "custom_schema",
      "origin": "catalog_authored",
      "definition_url": "https://github.com/magicsword-io/openattacks/blob/main/detections/analysis-review.yaml",
      "query_downloads": {
        "sigma": "/api/detections/analysis-review.sigma.yml",
        "kql": "/api/detections/analysis-review.kql",
        "spl": "/api/detections/analysis-review.spl"
      },
      "validation": "draft",
      "evidence": [
        "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
      ],
      "reviewed": "2026-10-23",
      "upstream_matches": []
    },
    {
      "node_id": "step-4",
      "profile": "analysis-review",
      "title": "Independent malware-analysis validation",
      "hypothesis": "Instructions inside a sample influence an LLM analysis result, or obfuscation changes a model's classification without changing malicious behavior.",
      "visibility": "Review / research workflow",
      "telemetry": [
        "Sample hashes and static / dynamic analysis results",
        "Model verdicts with prompt provenance",
        "Independent analyst and deterministic scanner verdicts"
      ],
      "investigate": [
        "Compare model conclusions with independent behavioral evidence and a controlled evaluation set.",
        "Treat sample comments as data; measure disagreement and false negatives without following embedded instructions."
      ],
      "limitation": "This evaluates analysis robustness, not victim endpoint compromise. Side behaviors mentioned elsewhere in a report do not establish edges in this flow.",
      "note": "Compare model/scanner verdicts with known behavior on the evaluated set; this does not establish a separate victim intrusion.",
      "queries": {
        "sigma": "title: Hunt - Independent malware-analysis validation\nid: aa44bd2e-f46c-4884-a6db-e6b370a12c50\nstatus: experimental\ndescription: Instructions inside a sample influence an LLM analysis result, or\n  obfuscation changes a model's classification without changing malicious\n  behavior.\nauthor: MagicSword OpenAttacks catalog\ndate: 2026-10-23\nfalsepositives:\n  - Authorized activity matching this broad hunt; baseline and investigate\n    before alerting.\nlevel: low\nlogsource:\n  category: application\n  product: openattacks_audit\n  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,\n    Target and CorrelationId from the required source logs first.\ndetection:\n  selection:\n    EventType:\n      - analysis.verdict\n  condition: selection\n",
        "kql": "// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.\n// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).\n// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.\nOpenAttacksAudit_CL\n| where TimeGenerated > ago(7d)\n| where EventType in (\"analysis.verdict\")\n| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId\n| order by TimeGenerated desc",
        "spl": "sourcetype=\"openattacks:audit\"\nEventType IN (\"analysis.verdict\")\n| table _time EventType Actor Asset Action Target CorrelationId\n| sort - _time"
      },
      "query_scope": "Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.",
      "query_kind": "custom_schema",
      "origin": "catalog_authored",
      "definition_url": "https://github.com/magicsword-io/openattacks/blob/main/detections/analysis-review.yaml",
      "query_downloads": {
        "sigma": "/api/detections/analysis-review.sigma.yml",
        "kql": "/api/detections/analysis-review.kql",
        "spl": "/api/detections/analysis-review.spl"
      },
      "validation": "draft",
      "evidence": [
        "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
      ],
      "reviewed": "2026-11-04",
      "upstream_matches": []
    }
  ],
  "attack_flow": {
    "nodes": [
      {
        "id": "step-1",
        "title": "Researchers provide malicious JavaScript",
        "description": "The experiment starts with existing code rather than an observed delivery campaign.",
        "kind": "access",
        "source_urls": [
          "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
        ]
      },
      {
        "id": "step-2",
        "title": "LLMs rewrite and obfuscate the code",
        "description": "Model output changes the script while retaining malicious behavior.",
        "kind": "ai",
        "source_urls": [
          "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
        ]
      },
      {
        "id": "step-3",
        "title": "Researchers test detection outcomes",
        "description": "The study compares detection before and after rewriting.",
        "kind": "execution",
        "source_urls": [
          "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
        ]
      },
      {
        "id": "step-4",
        "title": "Demo measures evasion",
        "description": "The finding concerns obfuscation efficacy, not a separate confirmed victim intrusion.",
        "kind": "impact",
        "source_urls": [
          "https://unit42.paloaltonetworks.com/using-llms-obfuscate-malicious-javascript/"
        ]
      }
    ],
    "edges": [
      {
        "source": "step-1",
        "target": "step-2"
      },
      {
        "source": "step-2",
        "target": "step-3"
      },
      {
        "source": "step-3",
        "target": "step-4"
      }
    ]
  }
}