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

Advisory record · `GHSA-935w-9g4m-p28p`

- **Identifiers:** `GHSA-935w-9g4m-p28p` · `CVE-2026-105752` · `PYSEC-2026-4308`
- **State:** Confirmed
- **Severity:** CVSS 3.1 Low
- **Critical:** No
- **Weakness (CWE):** [CWE-200](https://cwe.mitre.org/data/definitions/200.html) · [CWE-524](https://cwe.mitre.org/data/definitions/524.html)
- **Tags:** [Information disclosure](https://agentprime.io/advisories/tag/info-disclosure.md) · [CWE-524](https://agentprime.io/advisories/tag/cwe-524.md)
- **Published:** 6 Oct 2026
- **Updated:** 15 Oct 2026
- **Reviewed by a person:** No

Published by the GitHub Advisory Database at 6 Oct 2026, 00:02 UTC. agentprime has not analysed this yet, so whether it affects any app is not known.

## Affected packages

- `vllm` (PyPI): affected `<0.30.0`; fixed in `0.30.0`

## Machine-written summary

Summary written by Claude Haiku from the source advisory. Not reviewed by a person.

> vLLM versions before 0.30.0 are affected. When a GPT-OSS Harmony request on the Responses endpoint uses a built-in or MCP tool, the follow-up turns after each tool call drop the caller's cache_salt, so their prompt prefixes are cached in the shared unsalted namespace. An authenticated user on a shared deployment who can guess the post-tool history may be able to learn, from per-turn cached-token counts, whether another tenant processed that prompt. Upgrade to vLLM 0.30.0 or later, which contains the fix.

Written 1 Nov 2026.

## Not analysed yet

We haven’t published our analysis of this Advisory yet. Once a person has reviewed it, this page will say what’s vulnerable, what has to be true for your app to be affected, and how to check.

## Severity and exploitation

- 3.1 Low, CVSS 3.1: `CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N`, from CNA (GitHub_M)

- **CISA KEV:** Not listed.
- **EPSS:** 0.0022, the 12th percentile, on 5 Nov 2026: FIRST’s estimate of the chance of exploitation in the next 30 days.

## Timeline

- 31 Oct 2026, 14:08 UTC: First seen by agentprime
- 31 Oct 2026, 14:08 UTC: [Confirmed by OSV](https://osv.dev/vulnerability/GHSA-935w-9g4m-p28p)

## Sources

- [GitHub Advisory Database: GHSA-935w-9g4m-p28p](https://github.com/advisories/GHSA-935w-9g4m-p28p)
- [OSV: GHSA-935w-9g4m-p28p](https://osv.dev/vulnerability/GHSA-935w-9g4m-p28p)
- <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>
- <https://github.com/vllm-project/vllm/releases/tag/v0.30.0>

Facts on this page come from the GitHub Advisory Database (CC-BY 4.0), OSV and the other sources listed, as published. We don’t reproduce full advisory descriptions, which can contain exploit code.

## Check whether this affects you

A Match means an installed version is in the affected range above. Most don’t affect the app; agentprime hasn’t analysed this one yet.

- With the agentprime MCP server (<https://api.agentprime.io/mcp>), ask your agent to check the project: `check_lockfile` sends a lockfile, `check_packages` a list of installed packages, and both return the Matches and the upgrades that clear them. `get_advisory` returns this record. Setup: <https://agentprime.io/mcp.md>
- In a browser: <https://agentprime.io/check/> reads your lockfile in the browser, never uploading it.
- Whether it touches your code, with file and line: the Claude Code plugin, <https://agentprime.io/plugin.md>
- As JSON: <https://api.agentprime.io/v1/advisories/GHSA-935w-9g4m-p28p>, or `POST https://api.agentprime.io/v1/match` with your packages (<https://agentprime.io/api.md>)

---

The web page: <https://agentprime.io/advisories/GHSA-935w-9g4m-p28p/>. Everything for agents: <https://agentprime.io/llms.txt>.