NVIDIA Display Driver EUVD-2026-90115

Requires local access to a host running the affected NVIDIA GPU Display Driver for Linux with an already-authenticated unprivileged account (PR:L), plus the ability to interact with the GPU kernel-mode device interface (e.g., /dev/nvidia* ioctls). No user interaction is needed (UI:N) and no network path exists (AV:L), so remote exploitation is not possible. Exploitation is limited by needing a foothold on the machine, by the reliability challenges typical of kernel UAF exploitation (heap grooming and race timing), and by driver version - only affected GeForce, RTX/Quadro/NVS, and Tesla driver branches are impacted. The 'Guest Driver' designation makes the scenario most relevant to virtualized/vGPU and multi-tenant datacenter deployments where an untrusted low-privileged guest can reach the shared GPU driver.

This is a genuine, if not urgent, local-privilege-escalation risk. The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, base 7.8 HIGH) is internally consistent with the description: a use-after-free (CWE-416) in the NVIDIA Linux GPU driver kernel-mode layer, triggerable by an already-authenticated unprivileged local user (PR:L), with high confidentiality, integrity and availability impact reflecting the full range of stated outcomes (code execution, DoS, privilege escalation, information disclosure, data tampering). Because the attack vector is LOCAL (AV:L), this is not remotely exploitable and requires the attacker to already have code-execution as some user on the host - it is a privilege-escalation/local-attack primitive, not an initial-access vector. No EPSS score, no CISA KEV listing, and no public exploit code are present in the supplied data, so per the precision rules this is 'no public exploit identified at time of analysis'; there is also no SSVC decision data provided. Kernel UAFs in GPU drivers are historically weaponizable but non-trivial to exploit reliably (heap grooming, race timing), which the AC:L rating arguably understates for a UAF, though NVIDIA rates the trigger itself as low-complexity. Real priority is MODERATE and concentrated in multi-tenant contexts (the 'Guest Driver' tag and Tesla CPE point to virtualized/vGPU and datacenter deployments where an untrusted low-privileged guest sharing the GPU stack is exactly the threat model).

An unprivileged local user (or a low-privileged workload in a shared/virtualized GPU environment) issues crafted ioctl or device interactions to the NVIDIA kernel-mode driver that trigger a use-after-free on a freed kernel object. By grooming the kernel heap to reclaim the freed allocation with attacker-controlled data, they corrupt kernel memory to escalate to root, leak kernel memory, or crash the system. No public proof-of-concept is identified in the supplied data.