CVE-2026-73558
Received Received - Intake

Integer Overflow in vLLM Leads to Data Leakage

Vulnerability report for CVE-2026-73558, including description, CVSS score, EPSS score, affected products, exploitability, helpful resources, and attack-flow context.

Publication date: 2026-08-13

Last updated on: 2026-08-13

Assigner: GitHub, Inc.

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 Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-08-13
Last Modified
2026-08-13
Generated
2026-08-13
AI Q&A
2026-08-13
EPSS Evaluated
N/A
NVD
EUVD

Affected Vendors & Products

Showing 3 associated CPEs
Vendor Product Version / Range
vllm_project vllm 0.27.0
vllm_project vllm to 0.21.0 (exc)
vllm_project vllm From 0.21.0 (inc) to 0.27.0 (inc)

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-190 The product performs a calculation that can produce an integer overflow or wraparound when the logic assumes that the resulting value will always be larger than the original value. This occurs when an integer value is incremented to a value that is too large to store in the associated representation. When this occurs, the value may become a very small or negative number.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

This vulnerability is an integer overflow in vLLM's activation_kernels.cu that occurs during memory offset calculations. When processing large batches or hidden dimensions, the calculation blockIdx.x * 2 * d can exceed 32-bit integer limits, causing incorrect memory access. This leads to users in the same inference batch potentially receiving each other's input or output data.

Detection Guidance

Check vLLM version with pip show vllm. If version is below 0.27.0, the system is vulnerable. Monitor logs for unexpected data in inference outputs or memory access errors during batch processing.

Impact Analysis

If exploited, this vulnerability could allow an attacker to craft requests that, when processed in the same batch as a victim's request, receive partial or complete copies of another user's sensitive inference results. This could expose confidential or proprietary information processed by the model.

Compliance Impact

This vulnerability could lead to unauthorized data exposure, violating confidentiality requirements in GDPR and HIPAA. It may result in data breaches where personal or protected health information is leaked to unauthorized parties, potentially causing legal and regulatory penalties.

Mitigation Strategies

Upgrade vLLM to version 0.27.0 or later immediately. Avoid processing sensitive requests in shared batches until updated. Review inference batch configurations to limit large token dimensions or sequence lengths.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-73558. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart