NVIDIA vGPU Manager CVE-2026-47521

Requires local access to a Linux host running the NVIDIA vGPU Virtual GPU Manager, plus low-level privileges sufficient to invoke the kernel-mode vGPU interface (PR:L) - no user interaction is needed. The attacker must be able to send crafted input to the kernel-mode layer to trigger the out-of-bounds read. Limiting factors: the vulnerability is not remotely reachable (AV:L), it is confined to the Linux vGPU Manager (not Windows or general GPU display drivers), and it requires an existing foothold on the hypervisor host; only deployments actually running the vGPU Manager (virtualization/VDI/cloud-GPU hosts) are affected.

This is a locally-exploitable kernel-mode out-of-bounds read (CWE-125) in NVIDIA vGPU Virtual GPU Manager for Linux. The vendor CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, 7.8 HIGH) is internally consistent with the description and I agree with it: exploitation requires local access and low privileges, with no user interaction. There is no EPSS score, no CISA KEV listing, and no public POC provided, so all exploitation signals point to theoretical/unproven risk rather than active threat. The 'RCE' and 'Buffer Overflow' tags overstate the issue - the CVSS vector is AV:L (local, not remote/network) and the root cause is an out-of-bounds READ, not a classic overflow write; the full C:H/I:H/A:H impact set including 'code execution' and 'data tampering' is the vendor's worst-case aggregate for an OOB read and is optimistic for a read primitive. No SSVC decision data was supplied. Net: a real but medium-priority host-side flaw whose significance is elevated in multi-tenant vGPU/VDI/cloud-GPU environments (a guest-adjacent or co-tenant local attacker on the hypervisor host), but not an internet-facing or mass-exploitable issue.

A low-privileged local user on a Linux hypervisor host running the NVIDIA vGPU Virtual GPU Manager issues crafted requests to the kernel-mode vGPU component, triggering an out-of-bounds read. The attacker uses the leaked kernel memory (and potential state corruption) to disclose sensitive data, crash the host (DoS), or chain toward privilege escalation. No public exploit code is identified at time of analysis, and exploitation is constrained by the requirement for prior local access.