CVE-2025-46570: vllm Package Suffers from Timing Attack Vulnerability
Summary
CVE-2025-46570: vLLM is an inference and serving engine for large language models (LLMs). Affects vllm across 2 version ranges.
< 0.9.0
< 77073c77bc2006eb80ea6d5128f076f5e6c6f54f
Fixed in 77073c77bc2006eb80ea6d5128f076f5e6c6f54f. 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 inference and serving engine designed for large language models (LLMs). It optimizes the processing of prompts and enhances performance through its PageAttention mechanism, which identifies and utilizes matching prefix chunks to speed up the prefill process. However, this optimization can lead to vulnerabilities if timing differences are exploited.
Vulnerability Detail
The vulnerability arises from the PageAttention mechanism in vLLM, which, prior to version 0.9.0, could be exploited through timing attacks. When a new prompt is processed, if a matching prefix chunk is found, the Time to First Token (TTFT) is significantly reduced. This timing difference can be recognized and exploited by an attacker, potentially leading to unauthorized access or manipulation of the model's responses.
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
77073c77bc2006eb80ea6d5128f076f5e6c6f54f
This is the first release that contains the fix.
pip install --upgrade "vllm==77073c77bc2006eb80ea6d5128f076f5e6c6f54f"Tip — Not sure which version your app runs? Vulert can match this fix to it. Scan free →
Workaround
As a temporary workaround, you can avoid using the PageAttention mechanism until you update to the patched version. However, this may impact the performance of the vllm package.
Conclusion
The CVE-2025-46570 vulnerability in the vllm package highlights the importance of timely updates and awareness of potential security risks associated with performance optimizations. Users are encouraged to upgrade to version 0.9.0 or later to mitigate this issue.
See every vulnerability recorded for vllm →
FAQ
CVE-2025-46570 is a vulnerability in the vllm package that allows attackers to exploit timing differences in the PageAttention mechanism, potentially leading to unauthorized access.
You can use the Vulert playground at Vulert Scanner to check if your application is affected by this vulnerability without needing to sign up.
If you cannot update the package right away, consider disabling the PageAttention mechanism as a temporary workaround to mitigate the risk.
Failing to fix this vulnerability may expose your application to timing attacks, which could lead to unauthorized access or manipulation of the model's outputs.
More information about vllm can be found on its GitHub repository at https://github.com/vllm-project/vllm.
References
Related vulnerabilities
Recent vulnerabilities recorded for vllm.
