GPUStack authentication bypass CVE-2026-102895 under active exploitation

GPUStack versions prior to 2.2.0 contain a trivial authentication-bypass vulnerability tracked as CVE-2026-102895 that is now seeing active exploit attempts. GPUStack is open-source software designed to manage clusters of GPUs, commonly deployed for AI and machine learning workloads. The flaw was remediated in GPUStack 2.2.0 released at the end of June, but the vulnerability was not publicly disclosed until the end of July, with formal CVE assignment following on November 4th.

The delayed public disclosure creates a dangerous exposure window for administrators who may not have associated the June release with a critical security fix. Systems running unpatched GPUStack instances remain vulnerable to unauthenticated attackers who can bypass authentication entirely, granting access to GPU cluster management interfaces and potentially the underlying AI workloads and data.

Vulnerability details and exposure timeline

• CVE-2026-102895: Trivial authentication-bypass vulnerability in GPUStack, classified as unauthenticated remote access to GPU cluster management.
• Affected versions: All GPUStack releases prior to 2.2.0; the fixed version 2.2.0 shipped at the end of June 2026.
• Disclosure gap: The security fix was released in June but the vulnerability was not made public until end of July, with CVE assignment delayed until November 4th, creating a prolonged window for unpatched exploitation.
• Target attractiveness: GPUStack manages GPU clusters for AI workloads, making compromised instances valuable for compute hijacking, model theft, or lateral movement into AI pipelines.
• Attack vector: Remote and unauthenticated; exploitation requires no valid credentials, lowering the barrier for automated scanning and mass exploitation.

Why it matters

Organizations running open-source AI infrastructure face elevated risk from this vulnerability due to the common pattern of delayed security patching in self-managed GPU clusters. The exploitation of GPUStack specifically threatens the confidentiality and integrity of AI training pipelines and model weights, while also providing attackers with high-value compute resources for cryptomining or further attacks. Linux system administrators and MLOps engineers should prioritize verifying their GPUStack deployments and upgrading to 2.2.0 or later immediately.