NVIDIA GPU Driver EUVD-2026-90099

Exploitation requires the attacker to already have local, low-privileged access (a running process/account) on a system with the NVIDIA GPU Display Driver installed on Windows or Linux - matching AV:L and PR:L; no user interaction is needed (UI:N). The specific limiting factors are: (1) local access is mandatory, so remote-only attackers cannot trigger it directly; (2) the vulnerable code is in the driver's kernel-mode layer, so the attacker must be able to invoke the affected kernel interface (e.g., an ioctl/system call path) from the low-privileged context; and (3) the affected GPU product families are GeForce, RTX/Quadro/NVS, and Tesla with the vulnerable driver branch installed. Reaching escalation of privilege or code execution beyond information disclosure/DoS depends on successfully leveraging the out-of-bounds read primitive.

The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, base 7.8) frames this as a locally exploitable kernel-mode flaw requiring low-privileged access but no user interaction, with high impact across confidentiality, integrity, and availability. The root cause per CWE-125 and the description is an out-of-bounds READ from kernel heap memory - a read primitive most naturally yields information disclosure and denial of service; the vendor's claim of possible code execution, privilege escalation, and data tampering represents a best-case exploit outcome rather than a guaranteed one, so the I:H/A:H portion of NVIDIA's own rating is somewhat aspirational for a pure OOB read. No EPSS score, SSVC decision data, or POC indicator was provided, and the CVE is not listed in CISA KEV, so there is no public exploit identified at time of analysis and no evidence of active exploitation. Realistically this is a solid local privilege-escalation/info-leak candidate on a hugely deployed driver, but the local-only attack vector (AV:L requires an attacker to already run code on the host) meaningfully caps its urgency versus a remote flaw - it is a patch-in-normal-cycle item, elevated by the enormous install base, not an emergency. Key missing signals: EPSS, POC availability, and the specific affected/fixed driver branch versions (the NVIDIA advisory 5861 in product-security would carry these but version details were not supplied here).

A local attacker with low-privilege access to a Windows or Linux host running the NVIDIA GPU Display Driver (e.g., a standard user, a sandboxed process, or malware seeking escalation) invokes a crafted request to the vulnerable kernel-mode interface. The out-of-bounds read from kernel heap leaks adjacent kernel memory, which can be used to defeat KASLR and disclose sensitive data or, chained further, to corrupt kernel state for privilege escalation or a system crash. No public proof-of-concept has been identified, and the local attack vector means the attacker must already have code execution on the target.