Red Hat Security Advisory: Red Hat AI Inference 3.4.4 (spyre)

A flaw in vLLM within Red Hat AI Inference Server 3.4.4 enables server-side request forgery (SSRF) via crafted batch input JSON. This vulnerability allows attackers to coerce the batch runner to perform arbitrary HTTP/HTTPS requests from the server context, potentially bypassing network access controls and accessing internal services or metadata endpoints. Red Hat has published advisories confirming the vulnerability but currently does not provide a patch or mitigation that meets their criteria for ease of use, applicability, and stability. The vulnerability is tracked as CVE-2026-34753 with a Red Hat CVSS v3 base score of 5.4. Exploitation could lead to information disclosure and possibly further compromise of the host system.

An attacker with the ability to control batch input JSON can exploit this SSRF vulnerability to make the server send arbitrary HTTP/HTTPS requests. This can lead to unauthorized access to internal services, including cloud metadata endpoints and internal HTTP APIs, potentially resulting in information disclosure. The vulnerability may also allow attackers to bypass access controls such as firewalls by using the server as a proxy, increasing the attack surface. There is no known active exploitation in the wild at this time.

Red Hat currently does not offer an official patch or mitigation that meets their criteria for this vulnerability. Users should monitor Red Hat advisories for updates or fixes. Until a fix is available, consider restricting access to the batch input interface to trusted users only and limit network access from the AI Inference Server to sensitive internal endpoints where possible. Engage Red Hat support or a Technical Account Manager for tailored guidance.