NVIDIA AIStore Critical Vulnerability Allows Unauthenticated Token Revocation

The National Vulnerability Database has recorded a critical security flaw in NVIDIA AIStore for Linux, identified as CVE-2026-65173. This vulnerability allows unauthenticated remote attackers to submit crafted token-revocation lists, potentially leading to data tampering and denial of service.

According to the NVD record, the issue stems from missing authentication controls within the token-revocation endpoint. An attacker does not need valid credentials to exploit this weakness. By submitting maliciously crafted revocation lists, a remote adversary can disrupt normal operations. This action may result in unauthorized data modification or complete service unavailability for affected systems.

Technical Context and Severity

The National Vulnerability Database assigns a CVSS score of 9.1 to CVE-2026-65173, classifying it as Critical. This high severity rating reflects the ease of exploitation and the significant impact on system integrity and availability. The weakness is categorized under CWE-306, which denotes missing authentication for critical functions. This classification highlights the failure to verify user identity before executing sensitive operations.

The specific mechanism involves the token-revocation endpoint, which should be protected by strict access controls. Instead, the endpoint accepts inputs from any remote source without verification. This design flaw enables attackers to manipulate token states remotely. Such manipulation can invalidate legitimate user sessions or disrupt data access controls within the AIStore framework.

Affected Versions and Scope

NVIDIA AIStore framework versions 0 through 4.9 are affected by this vulnerability. Any deployment using these versions is potentially exposed to remote exploitation. The wide range of affected versions suggests that many existing installations may be vulnerable. Organizations using NVIDIA AIStore for Linux should verify their current version immediately.

The impact includes data tampering and denial of service. Data tampering refers to unauthorized changes to stored information, which can compromise data integrity. Denial of service occurs when legitimate users are prevented from accessing the system. Both outcomes pose significant risks to enterprises relying on NVIDIA AIStore for their data infrastructure.

What should you do?

Security teams should audit their NVIDIA AIStore deployments to determine if they are running versions 0 to 4.9. If affected versions are in use, isolate the systems from untrusted networks to reduce exposure. Monitor logs for unusual token-revocation activity or unexpected service interruptions. Contact NVIDIA support for guidance on mitigation or potential patches.