NVIDIA Display Driver EUVD-2026-90116

Requires local access to a system running the NVIDIA GPU Display Driver for Linux: the attacker must be an unprivileged local user (or, per the 'Guest Driver'/'Virtual Gpu Manager' tags, a guest using the NVIDIA vGPU/guest driver) able to open the driver's kernel device node and invoke ioctls. Exploitation is limited by the local attack vector (AV:L) - no remote or network path exists - but is otherwise unconstrained: no user interaction (UI:N), only low privileges (PR:L), low attack complexity (AC:L), and default driver installations expose the vulnerable kernel-mode ioctl interface. Highest-risk deployments are multi-tenant GPU environments (vGPU, VDI, cloud/HPC GPU nodes) where the flaw breaks user-to-kernel or guest-to-host isolation.

All signals point to a genuine but locally-scoped privilege-escalation risk rather than a remotely-wormable emergency. The CVSS 3.1 vector CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H (7.8, HIGH) confirms local attack vector, low complexity, and a low-privileged (authenticated local) caller with no user interaction, yielding high confidentiality, integrity and availability impact - consistent with a kernel out-of-bounds write (CWE-787). No EPSS score, no CISA KEV listing, and no public POC are present in the supplied data, so exploitation status is 'no public exploit identified at time of analysis'; this is not evidence of safety but reflects that no automated exploitation signal exists. The main risk-limiting factor is AV:L: an attacker must already be able to execute code on the host and open the NVIDIA kernel device node, so this is a local escalation primitive, not an internet-facing flaw. There is no conflicting data - CVSS local vector, the description ('unprivileged caller', 'kernel mode layer'), and the 'Guest Driver / Virtual Gpu Manager' tags all agree that the attacker is a local low-privileged user. Priority is elevated in multi-tenant contexts (vGPU/VDI, cloud GPU, shared HPC/ML nodes) where crossing from an unprivileged user to kernel is a tenant-isolation break; standalone single-user workstations are lower priority.

A local unprivileged user (or a guest VM using the NVIDIA guest/vGPU driver) issues a crafted ioctl to the NVIDIA kernel-mode driver with an input buffer whose size is not validated, triggering an out-of-bounds write in kernel memory. By controlling the overwritten kernel data, the attacker corrupts kernel structures to escalate to root/kernel privileges, or simply crashes the host for denial of service. No public exploit code is identified in the provided data, and exploitation requires prior local access.