Vulnerability-Lookup

GHSA-X9R9-C232-4Q39

Vulnerability from github – Published: 2026-09-01 15:31 – Updated: 2026-09-01 15:31

Details

A vulnerability in Hugging Face Transformers (versions 4.49.0, <= 5.8.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). The function fetches and caches a remote module file before performing the required trust_remote_code consent check, inverting the security model enforced by other code-loading paths (such as AutoConfig, AutoModel, and AutoTokenizer). As a result, attacker‑controlled Python code from custom_generate/generate.py is copied into the user’s ~/.cache/huggingface/modules directory even if the user declines the trust prompt. Although execution is correctly gated, the file write is not reversible and can persist across sessions. This can lead to persistent, unauthorized files on disk and stale cache collisions where cached attacker code may later be executed during trusted model loads. The issue stems from an unconditional file write in dynamic_module_utils.py prior to any trust verification.

Show details on source website

JSON

To clipboard

{
  "affected": [],
  "aliases": [
    "CVE-2026-80047"
  ],
  "database_specific": {
    "cwe_ids": [],
    "github_reviewed": false,
    "github_reviewed_at": null,
    "nvd_published_at": "2026-09-01T14:17:41Z",
    "severity": null
  },
  "details": "A vulnerability in Hugging Face Transformers (versions 4.49.0, \u003c= 5.8.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). The function fetches and caches a remote module file before performing the required trust_remote_code consent check, inverting the security model enforced by other code-loading paths (such as AutoConfig, AutoModel, and AutoTokenizer). As a result, attacker\u2011controlled Python code from custom_generate/generate.py is copied into the user\u2019s ~/.cache/huggingface/modules directory even if the user declines the trust prompt. Although execution is correctly gated, the file write is not reversible and can persist across sessions. This can lead to persistent, unauthorized files on disk and stale cache collisions where cached attacker code may later be executed during trusted model loads. The issue stems from an unconditional file write in dynamic_module_utils.py prior to any trust verification.",
  "id": "GHSA-x9r9-c232-4q39",
  "modified": "2026-09-01T15:31:13Z",
  "published": "2026-09-01T15:31:13Z",
  "references": [
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-80047"
    },
    {
      "type": "WEB",
      "url": "https://github.com/huggingface/transformers"
    },
    {
      "type": "WEB",
      "url": "https://kb.cert.org/vuls/id/456290"
    },
    {
      "type": "WEB",
      "url": "https://www.kb.cert.org/vuls/id/456290"
    }
  ],
  "schema_version": "1.4.0",
  "severity": []
}

CVE-2026-80047 (GCVE-0-2026-80047)

Vulnerability from cvelistv5 – Published: 2026-09-01 13:40 – Updated: 2026-09-01 18:09

Title

Hugging Face Transformers library writes remote code to disk prior to consent check

Summary

A vulnerability in Hugging Face Transformers (versions >= 4.49.0 and <= 5.8.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). The function fetches and caches a remote module file before performing the required trust_remote_code consent check, inverting the security model enforced by other code-loading paths (such as AutoConfig, AutoModel, and AutoTokenizer). As a result, attacker‑controlled Python code from custom_generate/generate.py is copied into the user’s ~/.cache/huggingface/modules directory even if the user declines the trust prompt. Although execution is correctly gated, the file write is not reversible and can persist across sessions. This can lead to persistent, unauthorized files on disk and stale cache collisions where cached attacker code may later be executed during trusted model loads. The issue stems from an unconditional file write in dynamic_module_utils.py prior to any trust verification.

Severity

No CVSS data available.

CWE

Assigner

certcc mitre root

References

3 references

URL	Tags
https://github.com/huggingface/transformers
https://kb.cert.org/vuls/id/456290
https://www.kb.cert.org/vuls/id/456290

Impacted products

1 product

Vendor	Product	Version
Hugging Face	Transformers	Affected: 4.49.0 , ≤ 5.8.1 (custom)

Show details on NVD website

JSON

To clipboard

{
  "containers": {
    "adp": [
      {
        "providerMetadata": {
          "dateUpdated": "2026-09-01T14:08:14.152Z",
          "orgId": "af854a3a-2127-422b-91ae-364da2661108",
          "shortName": "CVE"
        },
        "references": [
          {
            "url": "https://www.kb.cert.org/vuls/id/456290"
          }
        ],
        "title": "CVE Program Container"
      }
    ],
    "cna": {
      "affected": [
        {
          "product": "Transformers",
          "vendor": "Hugging Face",
          "versions": [
            {
              "lessThanOrEqual": "5.8.1",
              "status": "affected",
              "version": "4.49.0",
              "versionType": "custom"
            }
          ]
        }
      ],
      "descriptions": [
        {
          "lang": "en",
          "value": "A vulnerability in Hugging Face Transformers (versions \u003e= 4.49.0 and \u003c= 5.8.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). The function fetches and caches a remote module file before performing the required trust_remote_code consent check, inverting the security model enforced by other code-loading paths (such as AutoConfig, AutoModel, and AutoTokenizer). As a result, attacker\u2011controlled Python code from custom_generate/generate.py is copied into the user\u2019s ~/.cache/huggingface/modules directory even if the user declines the trust prompt. Although execution is correctly gated, the file write is not reversible and can persist across sessions. This can lead to persistent, unauthorized files on disk and stale cache collisions where cached attacker code may later be executed during trusted model loads. The issue stems from an unconditional file write in dynamic_module_utils.py prior to any trust verification."
        }
      ],
      "problemTypes": [
        {
          "descriptions": [
            {
              "description": "CWE-273 Improper Check for Dropped Privileges",
              "lang": "en"
            }
          ]
        },
        {
          "descriptions": [
            {
              "description": "CWE-367 Time-of-Check Time-of-Use (TOCTOU Race Condition)",
              "lang": "en"
            }
          ]
        },
        {
          "descriptions": [
            {
              "description": "CWE-494 Download of Code Without Integrity Check",
              "lang": "en"
            }
          ]
        }
      ],
      "providerMetadata": {
        "dateUpdated": "2026-09-01T18:09:39.169Z",
        "orgId": "37e5125f-f79b-445b-8fad-9564f167944b",
        "shortName": "certcc"
      },
      "references": [
        {
          "url": "https://github.com/huggingface/transformers"
        },
        {
          "url": "https://kb.cert.org/vuls/id/456290"
        }
      ],
      "source": {
        "discovery": "UNKNOWN"
      },
      "title": "Hugging Face Transformers library writes remote code to disk prior to consent check",
      "x_generator": {
        "engine": "VINCE 3.0.45",
        "env": "prod",
        "origin": "https://cveawg.mitre.org/api/cve/CVE-2026-80047"
      }
    }
  },
  "cveMetadata": {
    "assignerOrgId": "37e5125f-f79b-445b-8fad-9564f167944b",
    "assignerShortName": "certcc",
    "cveId": "CVE-2026-80047",
    "datePublished": "2026-09-01T13:40:05.952Z",
    "dateReserved": "2026-08-25T17:29:08.429Z",
    "dateUpdated": "2026-09-01T18:09:39.169Z",
    "state": "PUBLISHED"
  },
  "dataType": "CVE_RECORD",
  "dataVersion": "5.2"
}

Sightings

Author	Source	Type	Date	Other

Nomenclature

Seen: The vulnerability was mentioned, discussed, or observed by the user.
Confirmed: The vulnerability has been validated from an analyst's perspective.
Published Proof of Concept: A public proof of concept is available for this vulnerability.
Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
Not confirmed: The user expressed doubt about the validity of the vulnerability.
Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.

Detection rules are retrieved from Rulezet.

The MITRE ATT&CK techniques below are AI-generated suggestions, inferred from the description of the vulnerability by the CIRCL/vulnerability-attack-technique-classification-roberta-base model, served locally by ML-Gateway. They have not been verified by an analyst and are provided for guidance only.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.

Action not permitted

GHSA-X9R9-C232-4Q39

CVE-2026-80047 (GCVE-0-2026-80047)

Tags

Sightings

Nomenclature