CVE-2026-44223vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters
- 6.5MediumCVSS base score
- 0.37%Exploitation in 30 daysEPSS · top 69% of all CVEs
- NoKnown exploitednot in CISA's KEV catalog
- YesPublic exploitsa reference or SSVC says so
Description
vLLM is an inference and serving engine for large language models (LLMs). From 0.18.0 to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.
Public exploit
An exploit or proof of concept is public: anyone can try it against an unpatched system. 2 references of the record are tagged as an exploit. CISA's SSVC rates its exploitation: a proof of concept is public.
Show the exploits2
References tagged exploit2 references
- https://github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw (opens in a new tab)x_refsource_CONFIRMexploit
- https://github.com/vllm-project/vllm/pull/38610 (opens in a new tab)x_refsource_MISCexploit
Timeline
ID reserved
By GitHub_M
- 7 days later
· 7 days later
Published
- 41 days later
· 41 days later
Record updated
Changes after publication (a new score, a public exploit, EPSS crossing 1, 10 or 50 %, leaving KEV) are kept since November 2026; earlier ones are not known.
Scores
| Scored by | CVSS | Score | Vector |
|---|---|---|---|
| GitHub_MCNA (assigner) | 3.1 | Medium6.5 | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H |
| NVDNVD · Modified | 3.1 | Medium6.5 | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H |
- EPSS (exploitation in 30 days)
- 0.37% · top 69% of all CVEs (18 Nov 2026)
- CISA SSVC
- Exploitation: poc · Automatable: no · Technical impact: partial
The list uses the assigner's score when there is one, then CISA's, then NVD's: Medium 6.5.
Affected products
- affected >= 0.18.0, < 0.20.0
Affected packages
From reviewed advisories of the GitHub Advisory Database.
References
References2
- https://github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw (opens in a new tab)x_refsource_CONFIRMexploit
- https://github.com/vllm-project/vllm/pull/38610 (opens in a new tab)x_refsource_MISCexploit
Sources
The same CVE in the primary databases and aggregators, to check against.
- CVE.org record (opens in a new tab)
- CVE List JSON (GitHub) (opens in a new tab)
- NVD (opens in a new tab)
- FIRST EPSS (opens in a new tab)
- EUVD EUVD-2026-29800 (opens in a new tab)
- OSV.dev (opens in a new tab)
- GitHub Advisory GHSA-83vm-p52w-f9pw (opens in a new tab)
- CIRCL Vulnerability-Lookup (opens in a new tab)
Badge for a README or a ticket
[](https://cveledger.app/cve/CVE-2026-44223)Record last updated .