CVE-2025-24357: vllm-project vllm Deserialization of Untrusted Data
Vulnerability summary
CVE-2025-24357 is a high (CVSS 8.8) Deserialization of Untrusted Data vulnerability in vllm-project vllm. EPSS estimates a 0.87% chance of exploitation in the next 30 days, higher than 58% of scored vulnerabilities.
Advisory lineage
Upstream 0 · Downstream 0 · Related 1Timeline
- 27-01-2025PublishedVulnerability first disclosed.
- 12-02-2025Last modifiedVulnerability information updated.
Description
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.
CVSS metrics
3 scores · 3 sources · 1 version- CVE.orgCNA7.5HighCVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:HDiffers
- OSV · PypiEcosystem advisory · calculated8.8HighCVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
EPSS trend
Probability
0.87%
Chance of exploitation in the next 30 days
Percentile
58%
Versus all other CVEs
30-day change
+0.17 pp
From 0.70% on 20-09-2026
Known exploited
Compare KEV catalogsNot in any of the 3 tracked catalogs.
Public exploits
0 foundNo public exploits indexed.
Weaknesses
CWE · root causeAffected systems
PyPI1 product
- vllm< d3d6bb13fb62da3234addf6574922a4ec0513d04 | < 0.7.0
Vllm1 product
- vllm< 0.7.0
Vllm-project1 product
- vllm< 0.7.0
3 vendors · 3 products
References
8 links- github.com/vllm-project/vllm/security/advisories/GHSA-rh4j-5rhw-hr54X Refsource CONFIRMWEBAdvisoryVendor Advisory
- github.com/vllm-project/vllm/commit/d3d6bb13fb62da3234addf6574922a4ec0513d04X Refsource MISCWEBFIXPatch
- github.com/vllm-project/vllmPACKAGE