PT-2026-24113 · Vllm · Vllm
Racerz-Fighting
·
Published
2025-08-21
·
Updated
2026-03-09
·
CVE-2026-25960
CVSS v3.1
9.8
Critical
| Vector | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H |
Name of the Vulnerable Software and Affected Versions
vLLM versions prior to 0.15.1
vLLM version 0.17.0
Description
vLLM is an inference and serving engine for large language models (LLMs). A Server-Side Request Forgery (SSRF) protection mechanism implemented in version 0.15.1 can be bypassed in the
load from url async method. This bypass occurs due to inconsistent URL parsing behavior between the validation layer and the actual HTTP client. The validation layer uses urllib3.util.parse url() while the HTTP client uses aiohttp with the yarl library for URL parsing. These parsers handle backslash characters (``) differently, allowing an attacker to bypass the hostname allowlist check and potentially access arbitrary internal or external services, resulting in a full SSRF attack. The vulnerable component is located in the vllm/connections.py file, specifically within the load from url async function. The attack scenario involves providing a malicious URL such as https://httpbin.org@evil.com/.Recommendations
Versions prior to 0.15.1: Update to version 0.15.1 or later.
Version 0.17.0: Update to a newer version that addresses the inconsistent URL parsing behavior.
Exploit
Fix
SSRF
Found an issue in the description? Have something to add? Feel free to write us 👾
Weakness Enumeration
Related Identifiers
Affected Products
Vllm