Independent & educational platform Not a government body. No link or affiliation with any government organisation. Made for learning only. Test only systems you own or are authorised to test. Become a Partner or Sponsor → Full disclaimer

CVE-2026-93840

medium

vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids().

Vendor / product: Vllm — vllm

vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow concurrent requests to sample tokens outside their allowlists.

CVSS
5.3
5.3 (CVSS v3.1)
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:N
Severity source: NVD CVSS v3.1 (nvd@nist.gov)
Timeline (UTC)
Published: 2026-09-18 20:17
Modified: 2026-09-28 18:36
NVD status: Analyzed
Last synced 3 days ago
Verified against the source API 2026-10-27 06:54 UTC
Exploitation
Not listed in CISA KEV (as of last catalog sync)
Absence from KEV does not mean a CVE is not exploited. Exploitation status: unknown (no authoritative evidence in our sources).

Affected products

vllm / vllm <0.29.0

Weakness classification (CWE)

Source: NVD / CISA. “NVD-CWE-Other” and “NVD-CWE-noinfo” mean NVD could not assign a specific weakness.

Mitigation & advisories

Patch / mitigation: https://github.com/vllm-project/vllm/commit/5b0e5b69ac1a3884a6479c9537789c95263cc804
Patch / mitigation: https://github.com/vllm-project/vllm/pull/49080

References

Data sources

NVD CVE API 2.0 — fetched 2026-10-27 06:54 UTC · record ↗

Learn the basics

Understanding CVEs and CVSS Scores

Related CVEs