Now: Nvidia Flaws Expose AI Models, Critical Infrastructure Now: Nvidia Flaws Expose AI Models, Critical Infrastructure

Researchers are urging enterprises that rely on Nvidia GPUs for their AI workloads to ensure that systems are ed against critical security vulnerabilities in an NVIDIA toolkit for running GPU-accelerated containers. If exploited, the bugs can allow attackers to gain access to sensitive data, steal proprietary AI models, or create operational disruptions.

NVIDIA released an update last September to  CVE-2024-0132, a time-of-check time-of-use (TOCTOU) vulnerability that earned a CVSS rating of 9 out of 10, in the NVIDIA Container Toolkit.

However, after closer inspection, researchers from Trend Micro and Wiz separately discovered a secondary flaw that the did not mitigate, so some users, even on ed systems, would still be at risk.

Researchers at Trend Micro flagged what they deemed this "incomplete" fix for CVE-2024-0132 in a recent blog post and wrote the related bug allows denial-of-service (DoS). This may have created confusion among those who thought their systems were protected once the initial was applied, security experts say.

Keep reading his article in Dark Reading, a Data Center Knowledge partner site