NVIDIA GPU Driver CVE-2026-47524

Exploitation requires local code execution on the host already at a HIGH privilege level (PR:H) - i.e., the attacker must be an administrator/root-equivalent or otherwise elevated user able to invoke the NVIDIA kernel-mode driver interface directly; there is no remote or network path and no user-interaction vector. A vulnerable NVIDIA GPU Display Driver must be installed and loaded: affected are GeForce, RTX/Quadro/NVS, and Tesla lines on Windows or Linux running driver versions below the fixed branch builds (e.g., 616.56/616.92/615.71.09/610.x/595.91.07/582.78 depending on product and branch; note one sub-case limited to Maxwell, Volta, and Pascal architectures prior to 582.78). The PR:H requirement is the dominant limiting factor - it excludes unprivileged local users and all remote attackers, making this primarily a defense-in-depth / privilege-boundary concern rather than an initial-access vulnerability.

This is a genuine but moderate-priority local vulnerability, not a remotely wormable critical. The CVSS 3.1 vector (AV:L/AC:L/PR:H/UI:N/S:U) and 6.7 score are internally consistent and I agree with them: exploitation requires local access AND high privileges (PR:H) on the host, which sharply constrains the attacker population to those already holding administrative/elevated rights. The description lists an expansive impact set (code execution, DoS, privilege escalation, information disclosure, data tampering), but the underlying defect is a CWE-125 out-of-bounds READ - a read primitive far more reliably yields info-disclosure and DoS than full code execution, so the 'might lead to code execution' language and the 'RCE' tag are aspirational rather than demonstrated (and 'RCE' is a loose label given AV:L local vector, not network). No EPSS score, no CISA KEV listing, and no public POC were provided, meaning there is no signal of active or imminent exploitation; the PR:H barrier makes broad opportunistic exploitation unlikely. Net: real defect worth patching in the maintenance cycle, but not a drop-everything emergency. Data gaps: EPSS/KEV/POC status all absent (treat as 'none identified', not confirmed-safe).

An attacker who already holds elevated/administrative privileges on a Windows or Linux host with a vulnerable NVIDIA driver issues a crafted request to the kernel-mode driver interface, triggering an out-of-bounds read. The most reliable outcome is leaking adjacent kernel memory (info disclosure) or crashing the driver (DoS); code execution is claimed by the vendor but is a harder, less certain outcome for a read primitive. No public exploit code was identified at time of analysis.