CVE-2026-93989

vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

Published Sep 19, 2026 CVSS 4.0 · 2.3 Low EPSS < 1% KEV · Not listed
Description
vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of concurrent requests, causing different in-flight HTTP requests to return incorrect tokens.

Analysis

AI-generated

Technical impact

The vulnerability arises when vLLM does not validate the token indices supplied in the bad_words parameter against the model’s output width during sampling parameter updates. If an attacker provides indices that exceed the logits array bounds, the library writes beyond the allocated buffer, corrupting the logits of other in‑flight API calls. This corruption causes concurrent HTTP requests to return incorrect tokens, compromising the integrity of the model’s responses but not enabling arbitrary code execution.

Affected Systems

vLLM versions up to and including 0.29.0, distributed by the vllm‑project, are affected. The issue exists in the sampling_params and bad_words modules of the worker GPU sampling path. No other versions are listed as impacted.

Risk and Exploitability

The CVSS score of 2.3 indicates a low severity. The EPSS score of 0.00201 indicates an extremely low probability of exploitation, and the vulnerability is not listed in CISA’s KEV catalog, implying a low likelihood of widespread exploitation. Nonetheless, an attacker can craft a request with malformed bad_words indices to corrupt logits for other concurrent requests. The attack requires only the ability to send a request to the vLLM endpoint; no privilege escalation is needed. The primary risk is to service correctness rather than system compromise.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade vLLM to version 0.30.0 or later, which includes bounding checks for bad_words indices.
  • If an upgrade is not immediately possible, avoid using the bad_words feature or limit any supplied indices to the model’s vocabulary size and the maximum generation length.
  • Integrate code‑level checks in your deployment to verify that any bad_words index passed to SamplingParams.update_from_tokenizer() is within the logits width constraint.

Generated by OpenCVE AI on September 23, 2026 at 02:08 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}

Type Values Removed Values Added
First Time appeared Vllm-project
Vllm-project vllm
Vendors & Products Vllm-project
Vllm-project vllm
Type Values Removed Values Added
Description vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of concurrent requests, causing different in-flight HTTP requests to return incorrect tokens.
Title vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words
First Time appeared Vllm
Vllm vllm
Weaknesses CWE-129
CPEs cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
Vendors & Products Vllm
Vllm vllm
References
Metrics cvssV3_1

{'score': 3.1, 'vector': 'CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N'}

cvssV4_0

{'score': 2.3, 'vector': 'CVSS:4.0/AV:N/AC:H/AT:P/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N'}

Weaknesses
  • CWE-129

    Improper Validation of Array Index

  • CWE-787

    Out-of-bounds Write

Sources

Status
PUBLISHED
Assigner
VulnCheck
Published
Updated
Sep 22, 2026, 4:00 PM
Reserved
Sep 19, 2026, 10:55 AM
Status
Analyzed
Published
Sep 19, 2026, 11:17 PM
Modified
Sep 28, 2026, 6:38 PM
Severity
Low
Public date
Sep 19, 2026, 10:58 PM