5.3

CVE-2026-73558: vllm-project vllm Integer Overflow or Wraparound

CVSS: 5.3v3.1MediumAnalyzed
AliasesGHSA-7m6h-x95x-82q5PYSEC-2026-3935CGA-2rm6-8w7v-rrxh
Published 13-08-2026Last modified 13-08-2026

Vulnerability summary

CVE-2026-73558 is a medium (CVSS 5.3) Integer Overflow or Wraparound vulnerability in vllm-project vllm. EPSS estimates a 0.41% chance of exploitation in the next 30 days, higher than 34.3% of scored vulnerabilities. 1 public exploit is recorded.

Overall risk
31/100
0100
Medium · default modelHow is this calculated?
CVSS score
5.3/10
010
Medium · v3.1 (CVE.org)
EPSS score
0.41%
0%100%
Low · +0.09 pp in 30 days
KEVNot listedIn no KEV catalog
RansomwareNone recordedNo known use recorded
Public exploits1github

Timeline

  1. 13-08-2026
    PublishedVulnerability first disclosed.
  2. 13-08-2026
    Last modifiedVulnerability information updated.

Description

vLLM is an inference and serving engine for large language models. Prior to 0.27.0, an integer overflow in blockIdx.x * 2 * d in activation_kernels.cu can cause act_and_mul_kernel to consume another batched user's input, allowing a request processed in the same inference batch to receive a partial or complete copy of another user's inference result. This issue is fixed in version 0.27.0.

CVSS metrics

4 scores · 4 sources · 1 version

EPSS trend

Probability

0.41%

Chance of exploitation in the next 30 days

Percentile

34%

Versus all other CVEs

30-day change

+0.09 pp

From 0.33% on 20-09-2026

20-09-202620-10-2026
EPSS score
0.41%

Known exploited

Compare KEV catalogs

Not in any of the 3 tracked catalogs.

Public exploits

1 found
  • githubUnverified

    Vllm

    Maturity: poc

Weaknesses

CWE · root cause

Affected systems

Chainguard6 products
  • kserve-huggingfaceserver-cuda-13.3All versions (No fix recorded)
  • lmcache-cuda-12.8All versions (No fix recorded)
  • nemo-rl-cuda-13.0-vllmAll versions (No fix recorded)
  • py3.10-vllm-cuda-12.4All versions (No fix recorded)
  • py3.12-vllm-cuda-12.4All versions (No fix recorded)
PyPI1 product
  • vllm< 0.27.0
Vllm1 product
Vllm-project1 product

4 vendors · 9 products

References

10 links