AVID-2026-R1487

Description

Arbitrary Code Execution via Deserialization in huggingface/transformers (CVE-2024-3568)

Details

The huggingface/transformers library is vulnerable to arbitrary code execution through deserialization of untrusted data within the load_repo_checkpoint() function of the TFPreTrainedModel() class. Attackers can execute arbitrary code and commands by crafting a malicious serialized payload, exploiting the use of pickle.load() on data from potentially untrusted sources. This vulnerability allows for remote code execution (RCE) by deceiving victims into loading a seemingly harmless checkpoint during a normal training process, thereby enabling attackers to execute arbitrary code on the targeted machine.

Reason for inclusion in AVID: CVE-2024-3568 describes arbitrary code execution via deserialization in the HuggingFace Transformers library when loading a potentially malicious checkpoint, enabling remote code execution through pickle.load. This directly concerns an AI framework used in building/training/serving AI models, i.e., an AI software stack. The vulnerability resides in a component (transformers) used to build/deploy general-purpose AI systems, representing a software supply-chain risk related to dependencies/artifacts in AI pipelines. The report provides explicit vulnerability behavior (RCE via deserialization of untrusted data) and references, giving sufficient evidence for classification.

References

• NVD entry
• https://huntr.com/bounties/b3c36992-5264-4d7f-9906-a996efafba8f
• https://github.com/huggingface/transformers/commit/693667b8ac8138b83f8adb6522ddaf42fa07c125

Affected or Relevant Artifacts

• Developer: huggingface
• Deployer: huggingface
• Artifact Details:

Impact

AVID Taxonomy Categorization

• Risk domains: Security
• SEP subcategories: S0100: Software Vulnerability
• Lifecycle stages: L06: Deployment

CVSS

CWE

Other information

• Report Type: Advisory
• Credits:
• Date Reported: 2024-04-10
• Version: 0.3.3
• AVID Entry