Introduction
A missing authorization check in the kernel-mode layer of the NVIDIA GPU Display Driver lets a local user improperly access GPU resources. On both Windows and Linux, that can lead to privilege escalation, code execution, and data tampering. NVIDIA holds roughly 94% of the discrete GPU market, so the blast radius covers consumer gaming rigs, professional workstations, enterprise AI clusters, and cloud GPU instances.
Technical Information
CVE-2026-24190 is in the kernel-mode layer of the NVIDIA Display Driver for Windows and Linux. The root cause is classified as CWE-862: Missing Authorization. In practice, the code path that handles GPU resource access requests does not sufficiently check whether the requesting user context is allowed to access the specific resources it asks for.
Root Cause
When a user process asks the NVIDIA kernel driver for GPU resources (memory regions, compute contexts, or other GPU objects), the driver is expected to confirm that the caller has the right permissions. In vulnerable versions, that check is absent or incomplete. A local, unprivileged user can therefore craft requests that the kernel-mode layer processes without enforcing access controls, and gain access to GPU resources that should be restricted.
Attack Flow
The attack is local. An attacker with an unprivileged user account on a system running a vulnerable NVIDIA driver would follow a general path like this:
- The attacker identifies the installed NVIDIA driver version and confirms it is in the vulnerable range.
- They send GPU resource access requests through the driver's user-mode interface (such as IOCTL calls on Linux or equivalent mechanisms on Windows), targeting resources that belong to other processes, the kernel, or privileged GPU contexts.
- The kernel-mode layer does not enforce proper authorization on these requests, so the driver processes them and grants access to the requested GPU resources.
- Depending on the attacker's goal, the unauthorized access can be used for:
- Denial of service: monopolizing or corrupting GPU resources to crash the driver or starve other processes.
- Information disclosure: reading GPU memory that belongs to other users, other processes, or the kernel itself.
- Data tampering: writing to GPU buffers or state without authorization.
- Privilege escalation and code execution: using the kernel-mode context the driver runs in to escalate from user to kernel privileges.
Severity Context
NVIDIA's May 2026 bulletin records a CVSS base score of 7.8 (High). At the time of this analysis, the NVD entry was still marked "Undergoing Enrichment" and NVD had not yet published its own calculated CVSS vector.
Where the bug sits makes it worse. GPU drivers run at ring 0 on most systems, so an authorization bypass in the kernel-mode layer can turn a local user compromise into full system compromise. That is most concerning in:
- Multi-tenant environments where multiple users share the same physical GPU hardware
- Shared workstations where untrusted users may have local accounts
- Cloud GPU instances (vGPU deployments) where tenant isolation depends on the driver's access controls
Patch Information
NVIDIA has patched CVE-2026-24190 in its May 2026 GPU Display Driver Security Bulletin (ID 5821), first released on May 19, 2026 and revised through May 21, 2026. The fix adds proper authorization enforcement to the GPU resource access paths in the kernel-mode layer, so requests are validated before any resources are handed off. It ships as updated driver binaries. Because NVIDIA's GPU drivers are proprietary, there is no separate hotfix or source-level commit to reference.
Patched Windows Driver Versions
| Driver Branch | Minimum Patched Version |
|---|---|
| R595 | 596.36 |
| R580 | 582.53 |
| R535 | 539.72 |
NVIDIA also notes that OEM-distributed Windows driver versions 595.95, 592.13, 582.42, and 539.69 contain the equivalent security fixes. The bulletin explicitly lists the Windows R590 branch as addressing CVE-2026-24190 as well.
Patched Linux Driver Versions
| Driver Branch | Minimum Patched Version |
|---|---|
| R595 | 595.71.05 |
| R580 | 580.159.03 |
| R535 | 535.309.01 |
Affected Product Families
Every major NVIDIA product family gets the fix: GeForce, NVIDIA RTX / Quadro / NVS, and Tesla cards on both platforms. vGPU software guest drivers and Cloud Gaming guest drivers on Linux also include it in their updated releases (vGPU 20.1 / 19.5 / 16.14 and the April 2026 Cloud Gaming release).
Bulletin Revision Notes
The first release on May 19 contained errors, including a wrong reference to an R570 branch for Linux GeForce and swapped R590/R595 labels. Revision 2.0 on May 20 corrected these to the R535/R595 branches and their proper driver versions. Revision 3.0 on May 21 then corrected the vGPU and Windows updated-version entries. Organizations tracking the bulletin should confirm they are reading the latest revision.
Discovery Credit
Sihyun Roh and Byoungyoung Lee from Compsec, Seoul National University (SNU), are credited with the discovery.
Obtaining the Patch
Consumer users can download the patched drivers from the NVIDIA Driver Downloads page. Enterprise vGPU and Cloud Gaming customers must get their updates through the NVIDIA Licensing Portal. On May 22, 2026, Tenable published matching Nessus detection plugins (ID 316513 for Windows, ID 316514 for Linux), which automatically verify whether the patched driver version is deployed.
Interim Mitigations
If you cannot patch right away, minimize local user access on affected hosts, restrict unprivileged users from interacting with GPU devices, and watch for unusual GPU driver crashes or unexpected GPU activity consistent with denial of service or unauthorized access attempts.
Affected Systems and Versions
The vulnerability affects NVIDIA Display Drivers on Windows and Linux in these driver branches and product families:
Windows (vulnerable versions prior to):
- R595 branch: versions before 596.36
- R590 branch: consult the May 2026 bulletin for the specific fixed version in this branch family
- R580 branch: versions before 582.53
- R535 branch: versions before 539.72
Linux (vulnerable versions prior to):
- R595 branch: versions before 595.71.05
- R580 branch: versions before 580.159.03
- R535 branch: versions before 535.309.01
Affected product families:
- NVIDIA GeForce (consumer)
- NVIDIA RTX / Quadro / NVS (professional)
- NVIDIA Tesla (data center)
- vGPU software guest drivers (versions prior to vGPU 20.1, 19.5, and 16.14)
- Cloud Gaming guest drivers on Linux (versions prior to the April 2026 Cloud Gaming release)
Vendor Security History
NVIDIA runs a dedicated Product Security Incident Response Team (PSIRT) and publishes its vulnerability-handling policies on its PSIRT Policies page. The company has consistently issued quarterly GPU Display Driver security bulletins. Recent ones include the October 2025, January 2026, and May 2026 bulletins, each listing multiple CVEs with fixed versions for Windows and Linux drivers. That regular cadence indicates a mature disclosure and patching process, and organizations can plan their endpoint management workflows around these predictable release cycles.
References
- NVD: CVE-2026-24190
- CVE Record: CVE-2026-24190
- NVIDIA Security Bulletin: GPU Display Drivers, May 2026
- NVIDIA PSIRT Policies
- Tenable Nessus Plugin 316513 (Windows)
- Tenable Nessus Plugin 316514 (Linux)
- NVIDIA Security Bulletin: GPU Display Drivers, January 2026
- Gaming On Linux: NVIDIA Reveal More GPU Driver Security Flaws for May 2026
- CISA Known Exploited Vulnerabilities Catalog
- TechPowerUp: NVIDIA Discrete GPU Market Share Dominance Expands to 94%
