A critical vulnerability in the NVIDIA Container Toolkit allows attackers to escape containers and gain full control over the host system, posing a significant risk to AI applications in cloud and on-premise environments that rely on GPU resources. The flaw, tracked as CVE-2024-0132 (CVSS score 9/10), is a Time-of-Check Time-of-Use (TOCTOU) vulnerability affecting Toolkit versions up to 1.16.1 and GPU Operator up to 24.6.1, and occurs due to inadequate isolation between the containerized GPU and the host system. Attackers can exploit this vulnerability via maliciously crafted container images, granting them access to the host’s file system and enabling command execution through writable Unix sockets like “docker.sock” and “containerd.sock”, leading to code execution, data exfiltration, and potential full system takeover. According to Wiz Research, 33-35% of cloud environments using the vulnerable Toolkit are at risk, highlighting the severity of the issue. The vulnerability impacts multi-tenant environments, potentially exposing sensitive data and secrets from other applications sharing the same node or cluster. NVIDIA released patches in version 1.16.2 of the Container Toolkit and GPU Operator 24.6.2 on September 26, 2024, following Wiz’s report on September 1, 2024. CTIX analysts strongly advise users to upgrade, as technical details remain withheld to prevent exploitation while organizations implement mitigations.