Vulnerabilities

CISA Confirms Active Exploitation of GPUStack Flaw CVE-2026-102895: Patch Now Available

CISA added GPUStack flaw CVE-2026-102895 to its KEV catalog. The bug grants admin access without logging in; version 2.2.0 fixes it.

CISA Confirms Active Exploitation of GPUStack Flaw CVE-2026-102895: Patch Now Available

On November 9, CISA added CVE-2026-102895, a vulnerability in GPUStack, to its catalog of vulnerabilities exploited in real-world attacks. The flaw lets an attacker gain administrator privileges without providing any login credentials. According to the project’s security advisory, all versions prior to 2.2.0 are vulnerable, and the fix ships in release 2.2.0.

What’s newly confirmed

The new development is the CISA catalog listing, not the first public disclosure of the vulnerability. The GPUStack advisory was published earlier. Confirmation of active exploitation makes it more urgent to check your deployments, especially any reachable from untrusted networks.

GPUStack manages GPU resources and serves artificial intelligence models. The vulnerability stems from the software treating certain requests as local and automatically granting them administrator privileges. According to the vendor, external traffic can also be identified this way when it passes through a proxy server running on the same machine.

What’s the impact

An unauthorized administrator can modify users and models, create API keys and read access credentials. Abusing administrative functions can also lead to code execution on the machines that run workloads.

CISA set a November 12 remediation deadline for the U.S. federal agencies covered by its requirements. That deadline applies only to those agencies, but other organizations should treat it as a signal to review their exposure urgently. The catalog does not confirm that this vulnerability has been used in ransomware campaigns.

What to do now

  • Identify the version of every GPUStack deployment and upgrade vulnerable instances to 2.2.0 or later.
  • Restrict API access from untrusted networks.
  • If the upgrade has to wait, apply the workarounds described by the project, including enforcing authentication with force_auth_localhost=true.
  • Review your proxy server configuration and how it handles headers that specify the client’s address.
  • Audit accounts, API keys and workload configurations for unauthorized changes.

SecurHUB’s take

AI systems need the same identity and access controls as the rest of your production infrastructure. Penetration testing can help verify that the way an application is deployed doesn’t let attackers bypass its security controls.

Sources: CISA — KEV catalog, GPUStack — security advisory, CVERadar

AuthorSecurHUB Team

Read more

Related articles

Want to know if this affects your company?

A free consultation with one of our experts — no obligation, in full confidence.

Let's talk