NVIDIA GPU Driver EUVD-2026-90133

Requires the attacker to already have unprivileged local access (PR:L) to a Linux host running the affected NVIDIA GPU Display Driver / Virtual GPU Manager; no user interaction is needed (UI:N). Exploitation targets the kernel-mode driver's handling of read-only memory protection via an incorrect authorization check. Limiting factors: no network vector (AV:L means the attacker must run code locally), and value is highest on shared or multi-tenant systems (vGPU/VDI/cloud-GPU) where crossing the unprivileged-to-kernel boundary breaches tenant or user isolation. The vendor patch is available via NVIDIA bulletin 5861.

This is a genuine local privilege-escalation risk but not an emergency. The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U, C:H/I:H/A:H, base 7.8) describes a locally-exploitable flaw requiring an attacker to already have unprivileged code execution on the host - it is NOT remotely exploitable despite the 'RCE' tag, which is misleading given AV:L. There is no EPSS score, no CISA KEV listing, and no public POC, so the signal is: no public exploit identified at time of analysis. No SSVC decision data was provided. The CWE-863 incorrect-authorization root cause lets an unprivileged user defeat read-only memory protection and write to memory marked read-only in the kernel-mode layer, which realistically yields kernel-level code execution / privilege escalation - the highest-impact outcome in the stated list. Net: a high-CVSS local EoP that matters most on shared/multi-tenant Linux GPU hosts (vGPU/VDI/cloud) but lacks any active-exploitation or ease-of-exploitation signal that would push it to top priority. The 'Authentication Bypass' tag is loose - this is an authorization bypass (CWE-863), not authentication.

An attacker with an unprivileged local account (or code running in a low-privilege context) on a Linux system with the affected NVIDIA GPU driver issues crafted requests to the kernel-mode driver that bypass its authorization check on read-only memory, then writes to protected kernel memory to corrupt state and escalate to kernel privileges. On multi-tenant vGPU/cloud-GPU hosts this could let one tenant compromise the host or other tenants. No public exploit code is known, and exploitation requires prior local access.