vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle
Historical advisory recorded 23 Oct 2026 15:48:27 UTC. Read the latest advisory.
cve-program · CVE-2026-105752
PublishedvLLM 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.
Read the publisher's advisoryAffected versions, risk and source history
Related advisories and sources
Publishers link these advisories through the identifiers shown below. A shared identifier or date alone does not tell you whether your software is affected or whether the advisories describe the same cause.
CVE-2026-105752 · CVE overview and exploitation status
cve-program · CVE-2026-105752
- https://github.com/vllm-project/vllm/security/advisories/GHSA-935w-9g4m-p28p
- https://github.com/vllm-project/vllm/pull/50195
- https://github.com/vllm-project/vllm/pull/51818
- https://github.com/vllm-project/vllm/commit/6a2a2bb02b563b83f946012959fd3927984d072a
- https://github.com/vllm-project/vllm/releases/tag/v0.30.0
This historical advisory does not include today's related changes. Open the latest advisory to find current connections.