Hugging Face Transformers SEW convert_config 代码注入 远程代码执行漏洞

Successful exploitation of CVE-2025-14926 allows an attacker to execute arbitrary code on a system running a vulnerable Hugging Face Transformers instance. This could lead to complete system compromise, including data theft, modification, or destruction. The attacker needs to trick a user into converting a specially crafted checkpoint file. The code will execute in the context of the user running the Transformers code, potentially granting access to sensitive data or resources. This vulnerability shares similarities with other code injection flaws where unchecked user input is used to construct and execute code, potentially leading to a complete takeover of the affected environment.

Developers and researchers using Hugging Face Transformers for natural language processing tasks are at risk. Specifically, those who routinely download and convert checkpoint files from external sources or integrate Transformers into automated pipelines are particularly vulnerable. Shared hosting environments where multiple users have access to the Transformers installation are also at increased risk.

• python / transformers:

import subprocess
result = subprocess.run(['transformers', 'convert_config', '--input', 'malicious_checkpoint.json'], capture_output=True, text=True)
print(result.stderr)

• python / transformers: Monitor for unusual process execution related to the transformers package, especially when handling checkpoint files. • python / transformers: Check for modified or newly created files in the Transformers installation directory after checkpoint conversion. • python / transformers: Review system logs for errors or warnings related to checkpoint conversion or configuration loading.

1. 2025-12-23Disclosure

   disclosure

1. 已保留2025-12-18
2. 发布日期2025-12-23
3. 修改日期2025-12-24
4. EPSS 更新日期2026-03-23

The primary mitigation for CVE-2025-14926 is to upgrade Hugging Face Transformers to version 4.57.1 or later. If upgrading immediately is not feasible, carefully scrutinize any checkpoint files before conversion, especially those from untrusted sources. Consider implementing input validation and sanitization within your code to prevent the execution of malicious code. While a WAF or proxy cannot directly prevent this vulnerability, they can be configured to detect and block suspicious checkpoint file uploads or conversions. There are no specific Sigma or YARA rules available at this time, but monitoring for unusual process execution related to Transformers is recommended.