CVEFinder.io

CVE-2026-105752

ℹī¸ low
🔍 Scan for this CVE
Summary

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, Harmony tool continuations submitted through "POST /v1/responses" requests rebuild the next-turn engine input without preserving the cache_salt value, placing the continuation prefix in the global unsalted cache namespace even when the caller enabled salting. On deployments with prefix caching enabled, which is the default, an authenticated tenant who can reconstruct a victim's low-entropy post-tool history can s

Description

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, Harmony tool continuations submitted through "POST /v1/responses" requests rebuild the next-turn engine input without preserving the cache_salt value, placing the continuation prefix in the global unsalted cache namespace even when the caller enabled salting. On deployments with prefix caching enabled, which is the default, an authenticated tenant who can reconstruct a victim's low-entropy post-tool history can submit the same continuation and use the cached_tokens_per_turn count to determine whether the prefix was previously processed, defeating the intended tenant isolation of salted prefix caching. This issue is fixed in version 0.30.0.

CVSS Score
3.1
Low
EPSS Score
0.2
Exploit Probability
Published Date
2026-10-05
First Seen: 2026-10-08
📊 Relative Risk Intelligence

This CVE is Lower Risk - more severe than 2.0% of all 365,616 vulnerabilities in our database.

#358,136
Below average severity
Severity Percentile
đŸŽ¯ CISA SSVC Assessment Updated: Oct 6, 2026
🔍 Exploitation Status
None
No known exploits
⚙ī¸ Automatable
NO
Requires human interaction
đŸ’Ĩ Technical Impact
Partial
Limited system impact
SSVC data provided by CISA
Last Modified 2026-10-08
Source NVD 🔗
CVSS Vector 3.1 CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N
CWE IDs (Weakness Types)
CWE-200 CWE-524

đŸ“Ļ Affected Products 1

Vendor Product Status Affected Versions
vllm vllm Affected
< 0.30.0

🔗 References 5

https://github.com/vllm-project/vllm/commit/6a2a2bb0...
Patch
https://github.com/vllm-project/vllm/pull/50195
Issue Tracking Patch
https://github.com/vllm-project/vllm/pull/51818
Issue Tracking Patch
https://github.com/vllm-project/vllm/releases/tag/v0...
Release Notes
https://github.com/vllm-project/vllm/security/adviso...
Mitigation Vendor Advisory

🔗 Related CVEs 6

CVE ID Severity CVSS EPSS Summary Published
CVE-2026-105753 đŸ”ļ medium 6.5 0.4 vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU ... 2026-10-05
CVE-2026-105754 đŸ”ļ medium 6.5 0.3 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the /inference/v1/generate endpoint ... 2026-10-05
CVE-2026-105755 đŸ”ļ medium 4.2 0.2 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at th... 2026-10-05
CVE-2026-105756 đŸ”ļ medium 6.5 0.3 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, OpenAI-compatible request models acc... 2026-10-05
CVE-2026-105757 đŸ”ļ medium 6.5 0.3 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, structured-output request failures c... 2026-10-05
CVE-2026-105758 đŸ”ļ medium 5.3 0.3 vLLM is an inference and serving engine for large language models. From 0.24.0 until 0.30.0, the Qwen2VLVideoBackend and... 2026-10-05
These CVEs affect the same products