All vulnerabilities

Servers and applications

CVE-2026-105752

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

Severity
Low 3.1
Published
28 Sept 2026
Last changed
8 Oct 2026
Added to UpHeron
3 Nov 2026
Fixed in
0.30.0
Exploit chance
0.2% in the next 30 daysHigher than 12% of CVEs. EPSS by FIRST, 19 Nov 2026
Weakness
Information exposure (CWE-200)CWE-524
Attack
Over the network · Only under certain conditions · Needs a user account · No user action needed
Attacker can
Change some data
CVSS vector
CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N
Also known as
GHSA-935w-9g4m-p28p
Sources
NVD, CVE Program, Project advisories (GitHub)

Description

On the GPT-OSS "Harmony" path (POST /v1/responses), a request that uses a built-in or MCP tool runs as a multi-turn loop: after each tool call vLLM re-renders the full next-turn Harmony prompt and re-submits it to the engine. Turn 1 correctly carries request.cache_salt, but the tool-continuation re-submission rebuilds the engine input via tokens_input(token_ids) with no cache_salt. The continuation prefix is therefore cached in the global unsalted namespace even though the caller opted into salting. A second tenant who can guess the low-entropy post-tool history submits the reconstructed continuation (unsalted) and reads exact per-turn cached-token counts from the Responses usage — restoring the prompt-membership oracle that cache_salt is documented to prevent.

Silently dropping a preserved salt after the supported tool workflow is enabled is a broken isolation control: the caller enabled salting and every turn should stay isolated, but continuation turns leak into the shared cache.

From vllm-project/vllm's security advisory. Shortened; the full text is there.

Affected software

SoftwareAffected versions
vLLMServers and applications · cna:vllm-project|vllm
  • Every version before 0.30.0
vLLMServers and applications · cpe:a:vllm:vllm
  • Every version before 0.30.0
vLLMServers and applications · github:vllm-project/vllm
  • Every version before 0.30.0

About vLLM

open-source software for large language model inference

From Wikidata, CC0 1.0.
Docker Hub pulls
44,195,225Docker Hub, 15 Nov 2026
GitHub stars
96,943GitHub, 15 Nov 2026
Forks
24,836GitHub, 15 Nov 2026
Latest release
v0.33.0rc3GitHub, 15 Nov 2026
Released
2 Nov 2026GitHub
License
Apache Software License 2.0Wikidata
First published on GitHub
9 Feb 2023GitHub

References

  • NVD https://nvd.nist.gov/vuln/detail/CVE-2026-105752
  • CVE.org https://www.cve.org/CVERecord?id=CVE-2026-105752
  • GitHub Advisory Database https://github.com/advisories/GHSA-935w-9g4m-p28p
  • github.com https://github.com/vllm-project/vllm/security/advisories/GHSA-935w-9g4m-p28p