VDB

CVE-2025-24357

CVE-2025-24357 PUBLISHED CVSS 7.5 HIGH

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

Reported by github_m · Published January 27, 2025

vllm is fixed in d3d6bb13fb62da3234addf6574922a4ec0513d04 and 0.7.0 (PyPI)

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.

Fixed versions

Upgrade to a fixed version or later. Language registries are listed first, then Linux distributions, then vendor products.

PackageEcosystemFixed in
vllmPyPId3d6bb13fb62da3234addf6574922a4ec0513d04, 0.7.0
tritonserver-backend-vllm-24.04Chainguard24.04-r9
vllmvllm0.7.0

Risk scores

CVSS 3.1
7.5
CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:H

Affected products

VendorProductVersions
vllm-projectvllm< 0.7.0
PyPIvllm0
chainguardtritonserver-backend-vllm-24.040
vllm-projectvllm0
chainguardpy3.10-vllm-cuda-11.80
chainguardpy3.10-vllm-cuda-12.60

Timeline

  • Jan 24, 2025 Fix PR Merged
  • Jan 27, 2025 CVE Published
  • Jan 27, 2025 Coalition ESS Score (10 in total, the first dated here)
  • Jan 28, 2025 EPSS Score (30 in total, the first dated here)
  • Feb 12, 2025 CVE Updated
  • Dec 20, 2025 Security Advisory

Query this with the API

The same record, from the CLI:

vulnetix vdb vuln CVE-2025-24357
vulnetix vdb fixes CVE-2025-24357

Or with curl, using the org id and key from your account page:

curl -s -H "Authorization: ApiKey ${VULNETIX_ORG_ID}:${VULNETIX_API_KEY}" \
  "https://api.vdb.vulnetix.com/v2/vuln/CVE-2025-24357/fixes/registry"

The Community plan is free and allows 50 queries a day. Get a free API key · API quickstart

Look up in the Resolve Platform →
$ Lookup Community · 50/day Open Lookup ›