The NVIDIA vGPU Virtual GPU Manager for Linux contains an out-of-bounds read flaw in its kernel-mode layer. This memory corruption can lead to a range of destructive outcomes: arbitrary code execution, denial of service, privilege escalation, sensitive information disclosure, and data tampering. The weakness is a classic boundary-check failure (CWE‑125).

The affected component is NVIDIA:Virtual GPU Manager on Linux platforms. No specific version numbers are disclosed in the available data, so all deployments of this module should be reviewed for the presence of the fix.

The CVSS base score is 7.8, reflecting substantial severity. The EPSS score is not available, and the issue is not listed in CISA's KEV catalog, which indicates no known public exploits yet. The flaw resides in the kernel-mode code, so an attacker would likely need to interact with the vGPU Manager process, possibly with elevated privileges or local access, to trigger the out-of-bounds read. If successfully exploited, the attacker could seize platform control or disrupt service.