Deserialization of Untrusted Data in vLLM - CVE-2025-24357
Published: January 27, 2025 / Updated: May 1, 2026
Vulnerability identifier: #VU128755
CSH Severity: Medium
CVSS v4: 7.5 [CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:A/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N]
CVE-ID: CVE-2025-24357
CWE-ID: CWE-502
Exploitation vector: Remote access
Exploit availability:
No public exploit available
Vulnerability details
The vulnerability allows a remote attacker to execute arbitrary code.
The vulnerability exists due to deserialization of untrusted data in hf_model_weights_iterator in vllm/model_executor/weight_utils.py when loading a model checkpoint downloaded from Hugging Face with torch.load. A remote attacker can supply a malicious model checkpoint to execute arbitrary code.
User interaction is required to fetch the pretrained repository remotely.
Affected software
vLLM
How to mitigate CVE-2025-24357
Install security update from vendor's website.
vLLM - update to 0.7.0