NVIDIA Display Driver CVE-2026-47604

Requires the attacker to already have local, authenticated low-privileged code execution on a host running the NVIDIA GPU Display Driver kernel-mode component (Windows or Linux). Exploitation is only meaningful where multiple security contexts share the same GPU - multi-user workstations, VDI, or cloud/virtual-GPU deployments - since the flaw lets one process read another process's GPU channel state. Limiting factors: no remote vector (AV:L), the attacker needs a pre-existing local foothold, and impact is confined to information disclosure (no write/integrity or availability effect); single-user desktops with one active context gain little from this primitive.

Signals point to a low-to-moderate, locally-scoped priority. The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N, base 5.5) describes a local, low-privileged, no-interaction read primitive limited to confidentiality (C:H, I:N, A:N) - there is no code execution, privilege escalation, or availability impact claimed by the vector, so the 'Authentication Bypass' tag overstates the mechanism (this is missing authorization / CWE-862, not auth bypass). No EPSS score, KEV listing, or POC is provided, so this is no public exploit identified at time of analysis; there is no evidence of active exploitation and SSVC would likely rate exploitation=none, automatable=no, technical impact=partial. This is a real but bounded issue: a genuine cross-process GPU memory/channel-state disclosure that matters most in multi-tenant, VDI, or cloud-GPU contexts where one local user reading another process's GPU channel state can leak sensitive data, but it is not a widespread-exploitation emergency. Note the CPE set lists only NVIDIA product families (geforce, rtx/quadro/nvs, tesla) while the description explicitly covers both Windows and Linux - the platform coverage in the description is broader than the tag set reflects.

On a shared or multi-tenant workstation, VDI host, or cloud-GPU instance, a local low-privileged user runs a crafted process that issues GPU driver calls to reference another process's GPU channel state. Because the kernel-mode driver omits the authorization check, the attacker reads memory/state belonging to a victim process, disclosing sensitive rendered or compute data. No user interaction or elevated privileges are required, but the attacker must already have local code execution on the host.