CVE-2026-105752

published 5 Oct 2026 · updated 6 Oct 2026

vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle

3.1LowCVSS 3.1EPSS 0%

No exploitation in the wild has been reported. EPSS puts the chance of exploitation in the next 30 days at 0%. No fixed version is recorded.

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.

Affected

Vendors, products and packages with version ranges

VendorProductVersionsFixed
vllm-project<0.30.0

Timeline

Every event on this vulnerability

Embed

A live severity badge for an advisory, README or ticket

CVE-2026-105752 severity on Zeptric[![CVE-2026-105752](https://zeptric.io/badge/cve/CVE-2026-105752.svg)](https://zeptric.io/cve/CVE-2026-105752)