NVIDIA Display Driver CVE-2026-24194

Attacker must already have local shell access (AV:L) with at least a low-privileged authenticated account (PR:L) on a Linux system running an affected NVIDIA Display Driver branch - GeForce/RTX/Quadro/NVS/Tesla before 535.309.01, 580.159.03, or 595.71.05, or a vGPU guest driver up to and including the March 2026 releases - and must be able to open NVIDIA character device nodes (/dev/nvidia*), which by default requires membership in the 'video' group or equivalent device ACL. No user interaction is needed (UI:N) and complexity is low (AC:L). The attack does NOT work remotely over the network, does NOT affect systems without the NVIDIA proprietary driver loaded (e.g., nouveau-only hosts), and is constrained on hardened multi-tenant systems where GPU device nodes are not exposed to untrusted users or containers.

The CVSS 3.1 vector AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H confirms a local, low-complexity, authenticated attack with high impact on all three CIA properties but no scope change. EPSS at 0.01% (1st percentile) and SSVC exploitation status 'none' both indicate no observed exploitation, and the CVE is not on CISA KEV. SSVC additionally rates the bug as not automatable but with total technical impact, consistent with a manual LPE rather than a wormable network flaw. Signals are internally consistent: high severity for any local user who already has a shell, low real-world urgency for internet-exposed risk. This is a meaningful priority on multi-tenant Linux hosts (shared HPC nodes, vGPU hypervisor guests, container hosts where GPU device nodes are exposed) and lower priority on single-user workstations.

An attacker with an existing low-privileged shell on a Linux box with the NVIDIA driver installed - for example a developer login on a shared GPU workstation, a CI runner with GPU pass-through, or a tenant VM running an affected vGPU guest driver - opens an NVIDIA device node and issues a crafted IOCTL that exercises the vulnerable kernel mode layer handler. Improper permission handling lets the syscall corrupt or bypass privilege checks, yielding root code execution or sensitive kernel memory disclosure. No public exploit identified at time of analysis, but the SSVC 'total' technical impact rating indicates a successful exploit yields full host compromise.