NVIDIA Display Driver EUVD-2026-90189

Requires local access to a system running the affected NVIDIA GPU Display Driver for Linux (GeForce, RTX/Quadro/NVS, Tesla branches, including the vGPU Guest Driver and Virtual GPU Manager components), with the attacker holding an unprivileged local account able to reach the kernel-mode driver's device interface - PR:L, AV:L, no user interaction (UI:N). Limiting factors: the vulnerability is not remotely reachable (no network vector), so an attacker must already have code execution or interactive access on the host or inside a vGPU guest; there is no user interaction requirement to lower the bar once local access exists. In multi-tenant vGPU deployments the Guest Driver/Virtual GPU Manager exposure means a guest-side attacker is a realistic threat model.

All signals point to a locally-exploitable privilege-escalation issue rather than a remotely-wormable one. 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) requires local access with low privileges but no user interaction, and yields full C:H/I:H/A:H impact - consistent with a local attacker who already has a foothold (any unprivileged user with access to the GPU device nodes) escalating to kernel-level code execution. No EPSS score, CISA KEV listing, or public POC was provided in the input, so there is no public exploit identified at time of analysis and no evidence of active exploitation; this is a vendor-reported (NVIDIA) issue disclosed through NVIDIA's product-security advisory, not a CISA-flagged one. The description's hedged language ('might not be preserved', 'might lead to') reflects the vendor describing a class of impacts rather than a demonstrated exploit chain. Net: this is a legitimate patch-worthy local EoP in a very widely deployed driver, but it is not an emergency remote-exploitation scenario. Priority is driven by the enormous install base and kernel-level impact, not by any exploitation-probability signal, all of which are absent.

A local unprivileged user on a Linux host with the NVIDIA GPU driver installed - or a tenant inside a vGPU guest - issues crafted ioctls/operations to the kernel-mode driver that cause read-only memory permissions to not be correctly preserved, turning a protected mapping into a writable one. By corrupting that now-writable kernel memory, the attacker executes code in kernel context and escalates from their low-privilege account to full system control. No public POC is referenced, and exploitation requires local access, so this is a post-foothold escalation rather than an initial-access vector.