NVIDIA GPU Driver EUVD-2026-90098

Requires local access to a host running the NVIDIA GPU Display Driver (Windows or Linux) on affected GeForce, RTX/Quadro/NVS, or Tesla hardware, plus an existing low-privileged authenticated context to issue calls into the kernel-mode layer - the attacker must already be able to run code on the machine. No user interaction is needed (UI:N), and no special non-default configuration is called out, but the local-only attack vector (AV:L) and required prior foothold (PR:L) are the key limiting factors: there is no remote or network-facing exploitation path, which substantially constrains real-world risk versus the 7.8 headline score.

This is a genuine local privilege/information-disclosure concern but not an urgent, mass-exploitable network threat. The CVSS vector (AV:L/AC:L/PR:L/UI:N/S:U) confirms exploitation requires local access and existing low-privilege authentication on the host - an attacker must already run code on the machine, which rules out remote drive-by exploitation. NVIDIA (the reporting vendor) rates it 7.8 High with C:H/I:H/A:H, but that reflects NVIDIA's standard boilerplate that an OOB read 'might lead to code execution, DoS, escalation of privileges, information disclosure, and data tampering'; the concretely described primitive is a kernel out-of-bounds READ (CWE-125) yielding kernel information disclosure, which is fundamentally a confidentiality primitive whose escalation to code execution/tampering is speculative and not demonstrated in the data provided. No EPSS score, no CISA KEV listing, and no public POC are present in the input, so all three exploitation-likelihood signals are absent - this points to a patch-on-normal-cycle issue rather than an emergency. No SSVC decision data was supplied. The main real-world weight comes from the extremely broad affected footprint (GeForce, RTX/Quadro/NVS, and Tesla datacenter GPUs across Windows and Linux), meaning a large attack surface even though each individual exploit is local-only. Treat as a meaningful local EoP/info-leak to patch on a standard cadence, not a same-day fire drill.

A low-privileged local user (or malware already running in a limited context) on a Windows or Linux host with a vulnerable NVIDIA GPU driver issues a crafted request to the kernel-mode driver that triggers an out-of-bounds read, leaking adjacent kernel memory. The disclosed kernel data (e.g., pointers defeating KASLR) can be used to strengthen a subsequent local privilege-escalation chain. No public exploit code is identified at time of analysis, and exploitation requires prior local access.