{
  "id": "pluto-loadmine-qwen2-malware-2026",
  "name": "Load/Mine malicious Qwen2 models on Hugging Face",
  "date_reported": "2026-11-04",
  "researcher": "Shaked Chen, Eitan Chassid, Yotam Perkal",
  "org": "Pluto Security",
  "sources": [
    {
      "url": "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/",
      "title": "Load/Mine: A malware campaign targeting Hugging Face and ModelScope",
      "publisher": "Pluto Security"
    }
  ],
  "magicsword_coverage": [],
  "borderline": false,
  "borderline_note": null,
  "ai_role": "ai_agent_or_mcp_as_target",
  "ai_role_secondary": "llm_generated_malware",
  "ai_role_detail": null,
  "technique_summary": "Malicious Qwen2 test models uploaded to Hugging Face (e.g. MENXIANGYU/qwen2-tiny) and mirrored on ModelScope. config.json auto_map field points to modeling_evil2.EvilModel which executes on trust_remote_code=True and vLLM versions before 0.14.0 without that flag (CVE-2026-22807). Payload runs fileless peakminer cryptominer via memfd_create, establishes .bashrc persistence, deploys reverse shells, and achieves GPUStack container escape. Models appear as legitimate Qwen2 variants for testing purposes.",
  "initial_access": "User loads malicious model from Hugging Face or ModelScope with trust_remote_code=True or vulnerable vLLM version",
  "ai_tooling_and_models": [
    "Hugging Face model repository",
    "ModelScope mirror",
    "Qwen2 model format",
    "vLLM <0.14.0 (CVE-2026-22807)",
    "Transformers library (trust_remote_code execution)"
  ],
  "other_tooling": [
    "peakminer cryptominer",
    "memfd_create (fileless execution)",
    ".bashrc persistence",
    "Reverse shell",
    "GPUStack container escape"
  ],
  "victims_and_sectors": "ML researchers, AI developers, organizations using Hugging Face models",
  "regions": [
    "Multiple regions"
  ],
  "attribution_or_cluster": "unknown",
  "scale_and_metrics": "Multiple malicious models uploaded including MENXIANGYU/qwen2-tiny; exact victim count unknown",
  "prevention_angle": "Pin and verify model checksums; avoid trust_remote_code=True; update vLLM to 0.14.0 or later; monitor memfd_create usage; scan .bashrc for unauthorized changes; detect outbound cryptocurrency mining traffic; review model config.json auto_map fields; use model scanning tools",
  "related_magicsword_projects": [],
  "source_notes": "Relevant detection rules:\n- Sigma: Suspicious Cron File Created (persistence via .bashrc) (https://github.com/SigmaHQ/sigma/blob/master/rules/linux/file_event/file_event_lnx_susp_cron_file_created.yml)\n- Sigma: Audio Capture via memfd_create (fileless execution pattern, can detect suspicious memfd_create usage) (https://github.com/SigmaHQ/sigma/blob/master/rules/linux/auditd/lnx_auditd_audio_capture.yml)\n- Note: Cryptominer detection typically requires network traffic analysis (cryptocurrency pool connections) and process monitoring for mining-specific CPU/GPU patterns. Container escape detection depends on specific container runtime audit logs.\n",
  "iocs": {
    "domains": [],
    "ips": [],
    "hashes": [],
    "urls": [
      "https://huggingface.co/MENXIANGYU/qwen2-tiny"
    ]
  },
  "last_reviewed": "2026-11-04",
  "confidence": "high",
  "region_notes": "Multiple regions reflects global Hugging Face and ModelScope user base",
  "atlas_techniques": [
    {
      "id": "AML.T0010.003",
      "rationale": "Attackers poisoned publicly available ML models on Hugging Face and ModelScope platforms",
      "source_urls": [
        "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
      ]
    },
    {
      "id": "AML.T0058",
      "rationale": "Malicious code injected via model config.json auto_map field executes during model loading",
      "source_urls": [
        "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
      ]
    },
    {
      "id": "AML.T0011.000",
      "rationale": "Cryptominer achieves persistence through .bashrc modification",
      "source_urls": [
        "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
      ]
    },
    {
      "id": "AML.T0034",
      "rationale": "Fileless peakminer cryptominer runs via memfd_create to evade detection",
      "source_urls": [
        "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
      ]
    }
  ],
  "attack_flow": {
    "nodes": [
      {
        "id": "upload",
        "title": "Malicious models uploaded to repositories",
        "description": "Attacker uploaded trojanized Qwen2 test models to Hugging Face and ModelScope with malicious config.json auto_map fields",
        "kind": "access",
        "source_urls": [
          "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
        ]
      },
      {
        "id": "load",
        "title": "Victim loads malicious model",
        "description": "User loads model with trust_remote_code=True or via vulnerable vLLM <0.14.0 (CVE-2026-22807)",
        "kind": "ai",
        "source_urls": [
          "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
        ]
      },
      {
        "id": "execute",
        "title": "Malicious code executes during model load",
        "description": "modeling_evil2.EvilModel runs from config.json auto_map field during model initialization",
        "kind": "execution",
        "source_urls": [
          "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
        ]
      },
      {
        "id": "miner",
        "title": "Fileless cryptominer deployed",
        "description": "peakminer runs via memfd_create for fileless execution; establishes .bashrc persistence",
        "kind": "impact",
        "source_urls": [
          "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
        ]
      },
      {
        "id": "escalation",
        "title": "Additional payloads and container escape",
        "description": "Reverse shells deployed; GPUStack container escape achieved",
        "kind": "impact",
        "source_urls": [
          "https://pluto.security/blog/hugging-face-malware-campaign-load-mine/"
        ]
      }
    ],
    "edges": [
      {
        "source": "upload",
        "target": "load"
      },
      {
        "source": "load",
        "target": "execute"
      },
      {
        "source": "execute",
        "target": "miner"
      },
      {
        "source": "miner",
        "target": "escalation"
      }
    ]
  }
}