CVE-2026-73558: vLLM Cross-User Data Leak Vulnerability
Summary
CVE-2026-73558: vLLM: Cross-User Data Leak Vulnerability. Affects vllm versions < 0.27.0. Fixed in 0.27.0. How to fix ↓
Check if this vulnerability affects your application
Upload a manifest or SBOM and we'll show whether you're affected, plus every other risky package in the same scan.
No signup required · Results in about 60 seconds
Background
vLLM is an open-source library designed for efficient inference of large language models. It allows users to process multiple requests in batches, optimizing performance and resource utilization. However, due to a critical vulnerability, it is susceptible to cross-user data leakage, where one user's data can be inadvertently exposed to another user within the same inference batch.
Vulnerability Detail
The vulnerability arises from an integer overflow in the act_and_mul_kernel kernel, specifically in the computation of the expression blockIdx.x * 2 * d. This overflow can lead to the output of one user's request being incorporated into another user's response. When the batch size and sequence length are configured in a certain way, it can result in complete or partial data leakage from one user to another, posing a significant risk of unauthorized information disclosure.
How To Fix
Latest published version: 0.32.0 as of 22 Oct 2026
pip install --upgrade "vllm==0.32.0"Upgrading to it may include breaking changes; the version below is the minimum fixed release for this vulnerability.
If you need the minimum change
0.27.0
This is the first release that contains the fix.
pip install --upgrade "vllm==0.27.0"Tip — Not sure which version your app runs? Vulert can match this fix to it. Scan free →
Workaround
As a temporary workaround, avoid processing multiple user requests in the same inference batch until the package is updated to the patched version.
Conclusion
The vLLM library has a critical vulnerability that can lead to cross-user data leakage. It is essential for users to update to the latest version to protect against potential information disclosure. Utilizing tools like Vulert can help monitor and alert for such vulnerabilities in real-time.
See every vulnerability recorded for vllm →
FAQ
The vulnerability is a cross-user data leak caused by an integer overflow in the vLLM library, allowing one user's inference results to be exposed to another user.
You can use the Vulert playground at Vulert Scanner to check if your application is affected by this vulnerability without requiring any signup.
The vulnerability can lead to unauthorized access to sensitive information from other users' inference results, which can have severe privacy implications.
Yes, a patch is available in version 0.27.0 of the vLLM package. It is recommended to update to this version to mitigate the risk.
The CVSS score of this vulnerability is 5.3, indicating a moderate level of severity.
References
Related vulnerabilities
Recent vulnerabilities recorded for vllm.
