// For flags

CVE-2025-24357

vLLM allows a malicious model RCE by torch.load in hf_model_weights_iterator

Severity Score

7.5
*CVSS v3.1

Exploit Likelihood

*EPSS

Affected Versions

*CPE

Public Exploits

0
*Multiple Sources

Exploited in Wild

-
*KEV

Decision

-
*SSVC
Descriptions

vLLM is a library for LLM inference and serving. vllm/model_executor/weight_utils.py implements hf_model_weights_iterator to load the model checkpoint, which is downloaded from huggingface. It uses the torch.load function and the weights_only parameter defaults to False. When torch.load loads malicious pickle data, it will execute arbitrary code during unpickling. This vulnerability is fixed in v0.7.0.

*Credits: N/A
CVSS Scores
Attack Vector
Network
Attack Complexity
High
Privileges Required
None
User Interaction
Required
Scope
Unchanged
Confidentiality
High
Integrity
High
Availability
High
Attack Vector
Network
Attack Complexity
High
Authentication
None
Confidentiality
Complete
Integrity
Complete
Availability
Complete
* Common Vulnerability Scoring System
SSVC
  • Decision:-
Exploitation
-
Automatable
-
Tech. Impact
-
* Organization's Worst-case Scenario
Timeline
  • 2025-01-20 CVE Reserved
  • 2025-01-27 CVE Published
  • 2025-01-27 CVE Updated
  • 2025-01-28 EPSS Updated
  • ---------- Exploited in Wild
  • ---------- KEV Due Date
  • ---------- First Exploit
CWE
  • CWE-502: Deserialization of Untrusted Data
CAPEC
Affected Vendors, Products, and Versions
Vendor Product Version Other Status
Vendor Product Version Other Status <-- --> Vendor Product Version Other Status
Vllm-project
Search vendor "Vllm-project"
 Vllm
Search vendor "Vllm-project" for product "Vllm"
 < 0.7.0
Search vendor "Vllm-project" for product "Vllm" and version " < 0.7.0"
en
Affected