NVIDIA vGPU Manager CVE-2026-24201

Exploitation requires local access to the hypervisor host or to a guest VM running on a hypervisor with NVIDIA vGPU software installed and the Virtual GPU Manager active. The attacker must hold at minimum a low-privileged account (PR:L per CVSS vector - unauthenticated remote exploitation is not possible). High attack complexity (AC:H) indicates exploitation is not straightforward and likely depends on specific memory state or timing conditions within the vGPU Manager process. The affected deployment mode is specifically virtualized GPU environments using NVIDIA vGPU software - bare-metal NVIDIA GPU deployments without the vGPU Manager component are not affected by this CVE. Standard standalone workstation or server GPU driver installations are outside the blast radius. Exploitation is not automatable per SSVC, meaning mass-scanning or worm-style exploitation is not a realistic threat vector.

Multiple signals converge to indicate moderate-to-low immediate exploitation priority. The CVSS base score of 5.8 (Medium) is consistent with local-only access (AV:L), high attack complexity (AC:H), and mixed impact (C:L/I:L/A:H). The high availability impact is the most significant concern - a successful exploit could crash or destabilize the vGPU Manager on the host, affecting all guest VMs sharing the GPU. EPSS at 0.01% (2nd percentile) reflects very low community-observed exploitation probability. SSVC corroborates this: Exploitation status is none, Automatable is no (high complexity precludes automated mass exploitation), and Technical Impact is partial. There is no KEV listing and no public POC at time of analysis. Despite a relatively low CVSS score, environments running vGPU in multi-tenant or cloud hosting contexts face a meaningful confidentiality and availability risk because a compromised guest VM could potentially leverage this to impact co-resident tenants via the shared host manager. No signal conflict identified between CVSS vector and description - both imply a local-access prerequisite.

An attacker with a low-privileged account on a guest VM or on the hypervisor host crafts a malformed request or IOCTL sequence targeting the NVIDIA Virtual GPU Manager's input-handling routines, deliberately triggering an out-of-bounds write in host memory. The memory corruption may crash the vGPU Manager process - taking down GPU access for all co-resident guest VMs - or, under favorable memory layout conditions, may corrupt adjacent data structures to leak or modify sensitive information. No public proof-of-concept exists at time of analysis, and the high attack complexity (AC:H) means reliable exploitation requires precise knowledge of memory layout and timing.