High

Arbitrary Remote Code Execution via `_attn_implementation_internal` Config Injection in huggingface/transformers

Type: advisorySources: OSV

A critical remote code execution vulnerability exists in all versions of the HuggingFace transformers library prior to version 5.3.0. The vulnerability allows an attacker to craft a malicious `config.json` file containing the `_attn_implementation_internal` field set to an attacker-controlled HuggingFace Hub repository ID. When a victim loads this model using the standard `AutoModelForCausalLM.from_pretrained()` API, the library downloads and executes arbitrary Python code from the attacker's repository with the victim's full OS privileges. This issue arises due to unfiltered deserialization of configuration attributes, insufficient sanitization of internal fields, and unsandboxed execution of downloaded kernels. The vulnerability bypasses the `trust_remote_code` security mechanism, is invisible to the victim, and exploits the standard documented usage pattern, making it particularly severe. Users are advised to upgrade to version 5.3.0 or later to mitigate this issue.

Products: Not specified
Affected versions: Not specified
Fixed versions: Not specified
CVEs: CVE-2026-4372
Confirmed fixed CVEs: Not specified
Updated in this feed: 7 Nov 2026, 08:18 UTC · Revision 1

Source evidence

Arbitrary Remote Code Execution via `_attn_implementation_internal` Config Injection in huggingface/transformers

core-osv · Original source

Published 24 May 2026, 13:40 UTC

osv:CVE-2026-4372

Complete source record

{
  "id": "c522dc06f12c1cd3a09eabfab490fb8c87e2258e0ced4a221fafc523726af883",
  "url": "https://osv.dev/vulnerability/CVE-2026-4372",
  "method": "structured",
  "payload": {
    "kind": "advisory",
    "title": "Arbitrary Remote Code Execution via `_attn_implementation_internal` Config Injection in huggingface/transformers",
    "summary": "A critical remote code execution vulnerability exists in all versions of the HuggingFace transformers library prior to version 5.3.0. The vulnerability allows an attacker to craft a malicious `config.json` file containing the `_attn_implementation_internal` field set to an attacker-controlled HuggingFace Hub repository ID. When a victim loads this model using the standard `AutoModelForCausalLM.from_pretrained()` API, the library downloads and executes arbitrary Python code from the attacker's repository with the victim's full OS privileges. This issue arises due to unfiltered deserialization of configuration attributes, insufficient sanitization of internal fields, and unsandboxed execution of downloaded kernels. The vulnerability bypasses the `trust_remote_code` security mechanism, is invisible to the victim, and exploits the standard documented usage pattern, making it particularly severe. Users are advised to upgrade to version 5.3.0 or later to mitigate this issue.",
    "evidence": "osv:CVE-2026-4372",
    "products": [],
    "severity": "high",
    "structured": {
      "id": "CVE-2026-4372",
      "aliases": [
        "CVE-2026-4372",
        "GHSA-29pf-2h5f-8g72",
        "PYSEC-2026-2289"
      ],
      "metrics": [
        {
          "type": "CVSS_V3",
          "score": "CVSS:3.0/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H"
        }
      ],
      "affected": [
        {
          "ranges": [
            {
              "repo": "https://github.com/huggingface/transformers",
              "type": "GIT",
              "events": [
                {
                  "introduced": "0"
                },
                {
                  "fixed": "aad13b87ed59f2afcfaebc985f403301887a35fc"
                },
                {
                  "fixed": "a7f8e7ff37d87d1a1a0c8cf607971c607741452f"
                }
              ],
              "database_specific": {
                "extracted_events": [
                  {
                    "cpe": "cpe:2.3:a:huggingface:transformers:*:*:*:*:*:*:*:*",
                    "range": [
                      {
                        "introduced": "0"
                      },
                      {
                        "fixed": "5.3.0"
                      }
                    ],
                    "source": "CPE_RANGE"
                  },
                  {
                    "range": [
                      {
                        "fixed": "a7f8e7ff37d87d1a1a0c8cf607971c607741452f"
                      }
                    ],
                    "source": "REFERENCES_COMMIT"
                  }
                ]
              }
            }
          ],
          "versions": [
            "v4.56.1-Vault-Gemma-preview",
            "4.55.0-GLM-4.5V-preview-GLM-4.5V-preview",
            "4.55.0-GLM-4.5V-preview",
            "v4.53.2-Ernie-4.5-preview",
            "v4.53.2-modernbert-decoder-preview",
            "v4.52.4-Kyutai-STT-preview",
            "v4.52.4-VJEPA-2-preview",
            "v4.52.4-ColQwen2-preview",
            "v4.51.3-CSM-preview",
            "v4.51.3-GraniteMoeHybrid-preview",
            "v4.51.3-LlamaGuard-preview",
            "v4.51.3-D-FINE-preview",
            "v4.51.3-SAM-HQ-preview",
            "v4.51.3-BitNet-preview",
            "v4.51.3-InternVL-preview",
            "v4.51.3-Janus-preview",
            "v4.51.3-TimesFM-preview",
            "v4.51.3-MLCD-preview",
            "v4.51.3-Qwen2.5-Omni-preview",
            "v4.50.3-DeepSeek-3",
            "v4.49.0-Mistral-3",
            "v4.49.0-AyaVision",
            "v4.49.0-SigLIP-2",
            "v4.49.0-SmolVLM-2",
            "v4.33.1",
            "v4.16.0",
            "v4.15.0",
            "v4.14.0",
            "v4.13.0",
            "v4.12.0",
            "v4.11.0",
            "v4.10.0",
            "v4.9.0",
            "v4.8.0",
            "v4.7.0",
            "v4.6.0",
            "v4.5.0",
            "v4.4.0",
            "v4.3.0.rc1",
            "4.3.0.rc1",
            "v4.2.0",
            "v4.1.1",
            "v4.1.0",
            "v3.5.0",
            "v3.4.0",
            "v3.3.1",
            "v3.3.0",
            "v3.2.0",
            "v3.1.0",
            "v3.0.2",
            "v3.0.1",
            "3.0.1",
            "v3.0.0",
            "v2.11.0",
            "v2.10.0",
            "v2.9.1",
            "v2.9.0",
            "v2.8.0",
            "v2.7.0",
            "v2.6.0",
            "v2.5.1",
            "v2.5.0",
            "v2.4.1",
            "v2.4.0",
            "v2.3.0",
            "v2.2.2",
            "v2.2.1",
            "v2.2.0",
            "v2.1.1",
            "v2.1.0",
            "v2.0.0",
            "1.2.0",
            "1.1.0",
            "1.0",
            "v0.6.2",
            "v0.6.1",
            "v0.6.0",
            "v0.5.1",
            "v0.5.0",
            "0.5.0",
            "v0.4.0",
            "v0.3.0",
            "v0.2.0",
            "v0.1.2",
            "0.1.2"
          ],
          "database_specific": {
            "source": "https://storage.googleapis.com/cve-osv-conversion/osv-output/CVE-2026-4372.json"
          }
        }
      ],
      "modified": "2026-10-28T02:30:50.364887007Z",
      "provider": "osv",
      "references": [
        "https://huntr.com/bounties/1f693a6e-6836-4b8b-a0bd-ca036fba8884",
        "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/4xxx/CVE-2026-4372.json",
        "https://nvd.nist.gov/vuln/detail/CVE-2026-4372",
        "https://github.com/huggingface/transformers/commit/a7f8e7ff37d87d1a1a0c8cf607971c607741452f"
      ],
      "provider_metadata": {
        "cwe_ids": [
          "CWE-1066"
        ],
        "cna_assigner": "@huntr_ai",
        "osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/4xxx/CVE-2026-4372.json"
      }
    },
    "related_cves": [
      {
        "value": "CVE-2026-4372",
        "evidence": "/id"
      }
    ],
    "security_relevant": true,
    "severity_evidence": "/severity/0/score"
  },
  "source_id": "core-osv",
  "document_id": 518730,
  "published_at": "2026-05-24T13:40:40.995+00:00"
}