The vulnerability in vLLM (CVE-2026-22778) affects versions from 0.8.3 to before 0.14.1. When an invalid image is sent to the multimodal endpoint, the Python Imaging Library (PIL) throws an error that vLLM returns to the client, leaking a heap address. This information leak reduces the address space layout randomization (ASLR) entropy from approximately 4 billion guesses to about 8 guesses. This leak can be chained with a heap overflow vulnerability in the JPEG2000 decoder used by OpenCV/FFmpeg, enabling remote code execution. The vulnerability is classified under CWE-532 and CWE-209. Red Hat advisories confirm the vulnerability and note that it is fixed in vLLM version 0.14.1.

The vulnerability allows an attacker to obtain a heap address leak from error messages returned by the vLLM multimodal endpoint. This significantly reduces the effectiveness of ASLR, facilitating exploitation of a separate heap overflow vulnerability in the JPEG2000 decoder of OpenCV/FFmpeg. Successful exploitation can lead to remote code execution with high impact on confidentiality, integrity, and availability. The CVSS v3.1 score is 9.8 (critical), reflecting network attack vector, no privileges required, no user interaction, and high impact on confidentiality, integrity, and availability.

A fixed version of vLLM is available: version 0.14.1. Users should upgrade to vLLM 0.14.1 or later to remediate this vulnerability. No additional mitigations are indicated by the vendor advisories. Patch status is confirmed by Red Hat advisories linked in the vendor advisory content.