Denial of Service Vulnerability in vLLM Due to Unbounded Cache on Disk
Summary
CVE-2025-29770: vLLM denial of service via outlines unbounded cache on disk. Affects vllm versions < 0.8.0. Fixed in 0.8.0. 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 language model serving, providing features such as structured output and guided decoding. It utilizes the outlines library to support structured output, which includes an optional cache for compiled grammars stored on the local filesystem. This cache is enabled by default, potentially leading to resource exhaustion if misused.
Vulnerability Detail
The vulnerability arises from the outlines library used in vLLM, which unconditionally employs a cache for compiled grammars. This cache can be exploited by a malicious user who sends numerous short decoding requests with unique schemas, causing the cache to grow uncontrollably. The result can be a Denial of Service (DoS) when the filesystem runs out of space. This issue specifically affects the V0 engine of vLLM, while the V1 engine remains unaffected. The vulnerability is classified under CWE-770, indicating a potential for resource exhaustion.
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
0.8.0
This is the first release that contains the fix.
pip install --upgrade "vllm==0.8.0"Tip — Not sure which version your app runs? Vulert can match this fix to it. Scan free →
Workaround
There are no effective workarounds in existing versions of vLLM other than restricting untrusted access to the OpenAI compatible API server to prevent abuse.
Conclusion
In summary, the vLLM package has a significant Denial of Service vulnerability due to an unbounded cache on disk, which can be exploited by malicious users. Updating to version 0.8.0 or later is essential to mitigate this risk. Users should also consider restricting access to the API server to prevent potential abuse.
See every vulnerability recorded for vllm →
FAQ
The vulnerability is a Denial of Service (DoS) issue caused by an unbounded cache on disk used by the outlines library in vLLM.
The vulnerability affects the V0 engine of vLLM. The V1 engine is not impacted.
You can use the Vulert playground at Vulert Scanner to check if your application is affected by this vulnerability without needing to sign up.
The CVSS score of this vulnerability is 6.5, indicating a moderate severity level.
Currently, there is no built-in option to limit the cache size in existing versions of vLLM. It is recommended to disable the cache by default.
References
Related vulnerabilities
Recent vulnerabilities recorded for vllm.
