skulldc

CISA lists GPUStack loopback authentication bypass CVE-2026-102895 as exploited

Before 2.2.0, GPUStack treated requests that looked like they came from 127.0.0.1 as the built-in administrator by default; CISA now lists the flaw as exploited.

What CISA added

On 2026-11-09 CISA added CVE-2026-102895 to its KEV catalog of known exploited vulnerabilities. The entry is named “GPUStack Improper Authentication Vulnerability” and lists the weaknesses CWE-287, CWE-290 and CWE-346. Known ransomware campaign use is listed as Unknown. For US federal civilian agencies bound by BOD 22-01, the due date is 2026-11-12.

The NVD record was published on 2026-11-04 and has the status Analyzed. NVD describes GPUStack as “an open-source GPU cluster manager for AI model serving and GPU instance provisioning”. According to NVD, exploitation was observed in the wild against a 2.1.1 deployment.

How the bypass works

The flaw sits in how GPUStack decided who was calling. Prior to 2.2.0, get_current_user in gpustack/api/auth.py treated a loopback request source as the built-in administrator when force_auth_localhost was false, and false was the default.

That shortcut would be harmless if the source address could be trusted, but it could not. The client_ip_getter function accepted either a same-host reverse proxy peer or a client-supplied X-GPUStack-Real-IP loopback value. Either way, a remote request could appear to originate from 127.0.0.1 and be handled as the administrator, with no credentials at all.

The KEV and NVD descriptions list what that access allowed: user and model changes, persistent administrator API-key creation, disclosure of worker and cloud-provider secrets, and code execution through POST /v2/inference-backends/from-yaml. The persistent API key is the part that outlasts the bug. Access gained before an upgrade may still work after it, which is one reason the KEV required action points to CISA’s “Forensics Triage Requirements” as well as to patching.

Severity

The CVSS score in the record is a secondary one, credited in NVD to GitHub security advisories: CVSS 3.1 base score 9.8 CRITICAL, with the vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H. Read field by field, that is a network attack vector, low attack complexity, no privileges required, no user interaction, unchanged scope, and high impact on confidentiality, integrity and availability.

What to do

The issue is fixed in version 2.2.0. The NVD references point to the v2.2.0 release, the fixing commit, pull request 5346 and the advisory GHSA-gj9m-f8ph-pmhr in the gpustack/gpustack repository.

The KEV required action is to apply mitigations in accordance with vendor instructions, following CISA’s BOD 26-04 guidance on prioritizing security updates based on risk, and to discontinue use of the product if mitigations are unavailable. Stakeholders are responsible for evaluating each asset’s internet exposure.

For teams running GPUStack, a few checks follow from the descriptions:

  • Upgrade any instance older than 2.2.0, starting with ones reachable from outside the host.
  • Note whether force_auth_localhost was left at its default and whether a reverse proxy runs on the same host, since those are the conditions the bypass relied on.
  • Review administrator API keys, users, models and inference backends for entries nobody on the team created, and rotate the worker and cloud-provider secrets an exposed instance held.

Source: CVE-2026-102895 (NVD, CISA KEV)