CVE-2026-22778
vLLM leaks a heap address when PIL throws an error
Description
vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is fixed in 0.14.1.
Severity
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:HAffected Products
| Vendor | Product | Affected versions |
|---|---|---|
| vllm-project | vllm | >= 0.8.3, < 0.14.1 |
References
4 LinksRecord Details
Related Tool Categories
Tool categories that test for or protect against this vulnerability class.
CVE record data © The MITRE Corporation, used under the CVE Terms of Use. Exploited-in-the-wild data from the CISA Known Exploited Vulnerabilities catalog (CC0). Exploit prediction scores by FIRST EPSS (first.org/epss). This site is not endorsed or certified by MITRE, NVD, CISA, or FIRST.