GPUStack Vulnerability Allows Attackers to Bypass Authentication and Access Sensitive AI Resources
A critical vulnerability in the open-source GPUStack software has been exploited by attackers, allowing them to bypass authentication and access sensitive artificial intelligence (AI) resources. The vulnerability, identified as CVE-2026-102895, affects all versions of GPUStack prior to 2.2.0, which was released at the end of June.
GPUStack is a popular tool used to manage clusters of graphics processing units (GPUs), which are typically employed for running AI workloads. The software is designed to provide a secure and efficient way to manage these resources, but in this case, it has been found vulnerable to a common attack pattern. Proxies often add custom headers to requests, which can be trusted by the recipient. However, an attacker can spoof these headers, allowing them to bypass authentication and access control.
In the case of GPUStack, the vulnerable header is “X-Gpustack-Real-IP”. This header is intended to identify the IP address from which a request originates, but it also allows attackers to trick the software into thinking that requests come from a trusted source. Specifically, if the “X-Gpustack-Real-IP” header identifies the loopback address (127.0.0.1) as the source of the request, GPUStack will bypass authentication.
We have observed attempts to exploit this vulnerability in the wild, with attackers using a straightforward GET request that includes the spoofed “X-Gpustack-Real-IP” header. The requests we’ve seen are looking for sensitive AI-related resources, including models and messages.
The exploitation of this vulnerability matters because it can allow unauthorized access to sensitive data and systems. In an era where AI is increasingly being used in critical applications, the security of these systems must be a top priority. The fact that attackers have already begun exploiting this vulnerability underscores the need for users to update their GPUStack software immediately.
To protect yourself from this vulnerability, it’s essential to ensure that your GPUStack software is up-to-date. The latest version, 2.2.0, has been released and includes a fix for CVE-2026-102895. Additionally, be cautious when using proxies or allowing custom headers in your requests, as these can increase the risk of exploitation.
By staying informed about vulnerabilities like this one and taking proactive steps to secure our systems, we can reduce the risk of successful attacks and protect sensitive data from falling into the wrong hands.
Source: SANS ISC — 2026-11-12