Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessGain user access inside guest VM
-
DeliveryWrite crafted data to shared GSP trace buffer
-
ExploitWin TOCTOU race window
-
ExecutionTrigger improper host-side access
-
ImpactAchieve code exec / info-disclosure / DoS on host
Vulnerability Assessment
| Weakness |
CWE
ASVS 5.0
V15 - Secure Coding and Architecture TOCTOU prevention
|
| Exploitation Conditions | Requires the host to be running NVIDIA Virtual GPU Manager with vGPU assigned to guest VMs, and the attacker must already have user-level code execution inside a guest VM (PR:L, local guest context - not remote/network). Exploitation targets the GSP tracing component specifically and depends on manipulating a buffer shared between guest and host, then winning a time-of-check-to-time-of-use race (CWE-367), which is timing-dependent and not guaranteed on each attempt. Limiting factors: no network exposure, guest foothold required, race-window reliability, and relevance is confined to multi-tenant or untrusted-guest vGPU/VDI/cloud-GPU hosts; single-tenant hosts with trusted guests face far lower practical risk. |
| Impact | This is a genuine but locally-scoped priority rather than an internet-facing emergency. The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, 7.8 HIGH) describes a local, low-privilege attacker with full CIA impact - but the description clarifies the 'local' actor is a GUEST VM user reaching the host GPU System Processor (GSP) tracing component through a shared buffer, i.e. a guest-to-host boundary crossing that is the real concern in multi-tenant vGPU/VDI/cloud-GPU deployments. Notably the CWE is CWE-367 (TOCTOU race condition), which conflicts somewhat with the vendor's AC:L: race conditions typically require winning a timing window, so real-world exploitation is likely closer to AC:H (I assess ~7.0) and non-deterministic. No EPSS score, no CISA KEV listing, and no public POC are present in the provided data, so there is no evidence of active or imminent exploitation. The 'RCE' tag overstates reachability - this is a local guest-context flaw, not remote code execution. Net: patch on a normal-to-elevated cadence for shared/multi-tenant GPU hosts; deprioritize for single-tenant workstation vGPU with trusted guests. |
| Exploit ScenarioDerived | A malicious or compromised user inside a guest VM on a vGPU-enabled host writes crafted data into a buffer shared with the host's GSP tracing component and races the host's time-of-check/time-of-use window to induce improper access. Winning the race lets the guest corrupt or read host GSP state, potentially escalating from guest context toward host code execution, information disclosure, or denial of service affecting co-tenant VMs. No public exploit code is identified at time of analysis, and the TOCTOU nature makes reliable exploitation timing-dependent. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Sep 30, 2026
Improper access through a buffer shared between a guest VM and the host in NVIDIA Virtual GPU Manager's GPU System Processor (GSP) tracing component (CWE-367, time-of-check-to-time-of-use) can lead to code execution, privilege escalation, data tampering, denial of service and information disclosure on the affected host. Exploitation is local rather than network-facing: the attacker must already hold user-level code execution inside a vGPU-attached guest VM and must additionally win a timing-dependent race against the host-side tracing path (CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H per independent assessment). No public exploit code and no confirmed active exploitation were identified at time of analysis, so practical risk is concentrated on multi-tenant vGPU, VDI and cloud-GPU hosts that run untrusted guests rather than single-tenant hosts with trusted workloads.
RemediationAI
Apply the fixed NVIDIA Virtual GPU Manager driver release identified in NVIDIA security bulletin 5861 (https://github.com/NVIDIA/product-security/tree/main/2026/5861); no exact fix version is given in the available input data, so verify the patched driver branch on that advisory before upgrading, and plan a host-level maintenance window because the vGPU Manager update requires restarting or migrating guest VMs and will interrupt active vGPU sessions. Until the host is patched, the most effective compensating control is to remove the attack precondition entirely: do not assign vGPU instances to untrusted or low-trust guests, and keep untrusted tenants on separate physical hosts from sensitive workloads - the trade-off is reduced tenant density and higher infrastructure cost. Where the host driver exposes configurable tracing or diagnostic interfaces, restrict or disable guest-visible tracing/debug surfaces (this may reduce telemetry that would otherwise help detect abuse), and audit which guest accounts hold the ability to trigger tracing operations. Restrict vGPU assignment and host management to administrative accounts, monitor host and GSP logs for anomalous tracing activity, buffer-related faults or unexpected guest VM crashes as possible exploitation indicators, and treat this as a hardening priority for multi-tenant environments rather than an internet-facing emergency since there is no network attack surface and no public exploit identified at time of analysis.
Affected VersionsENISA
Affects your stack
Find out whether the stacks and servers you run are on an affected version.
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
NVIDIA Virtual GPU Manager is the host-side driver stack that mediates vGPU instances assigned to guest VMs, and the GPU System Processor (GSP) is the on-GPU microcontroller whose firmware handles scheduling, memory management and tracing services on behalf of the host driver. The vulnerability sits in the GSP tracing path: guest-supplied data passed through a buffer shared between guest and host is not properly validated or serialized, so the state checked by the host can change before it is used - the classic CWE-367 time-of-check-to-time-of-use race. Because the buffer is shared, a guest can craft contents that the host later consumes as trusted input, producing improper access to host/GSP resources. The affected product is recorded in CPE as nvidia:virtual_gpu_manager with all version fields wildcarded, meaning the source data did not enumerate specific vulnerable versions; CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H (independent assessment) reflects a local, timing-dependent attack requiring low privileges inside the guest, with high confidentiality, integrity and availability impact on the affected host subsystem.
Know when Virtual Gpu Manager gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H
Local guest foothold (AV:L, PR:L); CWE-367 TOCTOU race requires winning a timing window so AC:H rather than vendor's AC:L; full CIA impact on host GSP retained.
Primary rating from Vendor (nvidia).
CVSS VectorVendor: nvidia
DescriptionCVE.org
NVIDIA Virtual GPU Manager contains a vulnerability in the GPU System Processor (GSP) tracing component where a guest VM user may cause improper access by sending crafted data through a shared buffer. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, data tampering, denial of service, and information disclosure.
Articles & Coverage 1
Executive ViewAI
A privilege-escalation flaw in NVIDIA Virtual GPU Manager's host-side GSP tracing component could allow a user inside a guest VM to break out of the guest/host boundary and execute code, tamper with data, or read sensitive information on the physical host. This matters most to organizations running multi-tenant vGPU, VDI, or cloud-GPU infrastructure, where a single malicious or compromised tenant could potentially affect co-hosted workloads and other customers. NVIDIA has not released a fixed driver version, and no public exploit code or confirmed active exploitation has been identified at time of analysis.
No vendor-released patch identified at time of analysis. Interim mitigations: disable or restrict the GSP tracing feature on affected hosts where operationally feasible; enforce strict tenant isolation and only allow trusted, hardened guest images on vGPU hosts; migrate untrusted or high-risk tenants to single-tenant or non-vGPU infrastructure; monitor host and guest telemetry for anomalous privilege-escalation or tracing-related activity; coordinate directly with NVIDIA support for hotfix or mitigation guidance.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Information disclosure occurs when an application unintentionally exposes sensitive data that aids attackers in reconnaissance or directly compromises security.
Remote Code Execution represents the critical moment when an attacker successfully runs arbitrary code on a target system without physical access.
Denial of Service attacks render applications or systems unavailable by overwhelming resources or triggering failure conditions.
More in Virtual Gpu Manager
View allLocal privilege escalation and code execution in the NVIDIA Display Driver for Linux (GeForce, RTX/Quadro/NVS, Tesla, an
An out-of-bounds read in the kernel-mode layer of NVIDIA GPU Display Driver on Windows and Linux allows a local, authent
Local privilege escalation and potential code execution in NVIDIA GPU Display Driver for Windows and Linux arise from a
Local privilege escalation in the kernel-mode layer of the NVIDIA GPU Display Driver for Linux allows an unprivileged lo
Privilege escalation in NVIDIA's GPU Display Driver for Windows lets a local, low-privileged authenticated user bypass a
Local privilege escalation in NVIDIA Display Driver for Linux (GeForce, RTX/Quadro/NVS, Tesla, and Virtual GPU Manager b
Local privilege escalation in NVIDIA Display Driver for Windows (GeForce, RTX/Quadro/NVS, Tesla, and vGPU guest/manager
Incorrect resource transfer between spheres in the NVIDIA vGPU Virtual GPU Manager for Linux allows a locally positioned
An out-of-bounds write in the kernel-mode layer of NVIDIA vGPU Virtual GPU Manager for Linux lets a local attacker with
Local privilege escalation, code execution, denial of service, and information disclosure affect Linux hosts running NVI
Local privilege escalation in NVIDIA vGPU Virtual GPU Manager for Linux stems from an out-of-bounds read in firmware (CW
Local attackers with low privileges on a Linux host running NVIDIA vGPU Virtual GPU Manager can trigger an out-of-bounds
EUVD-2026-90059
GHSA-m29v-vmff-9v9r
NVD
CVE.org
EPSS
Vulnrichment
Exploit-DB