# vLLM has Remote DoS via Invalid Recovered Token Reinjection

Advisory record · `GHSA-8wr5-jm2h-8r4f`

- **Identifiers:** `GHSA-8wr5-jm2h-8r4f` · `CVE-2026-54234` · `PYSEC-2026-3542`
- **State:** Confirmed
- **Severity:** CVSS 7.5 High
- **Critical:** No
- **Weakness (CWE):** [CWE-20](https://cwe.mitre.org/data/definitions/20.html) · [CWE-1284](https://cwe.mitre.org/data/definitions/1284.html)
- **Tags:** [CWE-20](https://agentprime.io/advisories/tag/cwe-20.md) · [CWE-1284](https://agentprime.io/advisories/tag/cwe-1284.md)
- **Published:** 17 Jul 2026
- **Updated:** 10 Sept 2026
- **Reviewed by a person:** No

Published by the GitHub Advisory Database at 17 Jul 2026, 17:08 UTC. agentprime has not analysed this yet, so whether it affects any app is not known.

## Affected packages

- `vllm` (PyPI): affected `>=0.17.1, <0.24.0`; fixed in `0.24.0`

## Machine-written summary

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

> GHSA-8wr5-jm2h-8r4f: vLLM versions >=0.17.1 and <0.24.0 are affected. A remote client that can send normal public gRPC Generate and Abort requests with overlapping speculative-decoding workloads can make an out-of-vocabulary recovered token reach the drafting path and crash the shared engine worker with a GPU device-side assert, causing a denial of service for other clients. Upgrade to vLLM 0.24.0, 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

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

- **CISA KEV:** Not listed.
- **EPSS:** 0.0062, the 49th percentile, on 9 Nov 2026: FIRST’s estimate of the chance of exploitation in the next 30 days.

## Timeline

- 31 Oct 2026, 14:11 UTC: First seen by agentprime
- 31 Oct 2026, 14:11 UTC: [Confirmed by OSV](https://osv.dev/vulnerability/GHSA-8wr5-jm2h-8r4f)

## Sources

- [GitHub Advisory Database: GHSA-8wr5-jm2h-8r4f](https://github.com/advisories/GHSA-8wr5-jm2h-8r4f)
- [OSV: GHSA-8wr5-jm2h-8r4f](https://osv.dev/vulnerability/GHSA-8wr5-jm2h-8r4f)
- <https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f>
- <https://nvd.nist.gov/vuln/detail/CVE-2026-54234>
- <https://github.com/vllm-project/vllm/pull/44744>
- <https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca>
- <https://github.com/vllm-project/vllm>

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-8wr5-jm2h-8r4f>, or `POST https://api.agentprime.io/v1/match` with your packages (<https://agentprime.io/api.md>)

---

The web page: <https://agentprime.io/advisories/GHSA-8wr5-jm2h-8r4f/>. Everything for agents: <https://agentprime.io/llms.txt>.