Introduction
A numeric type conversion error in NVIDIA's Linux kernel-mode display driver can cause a heap buffer overflow, which gives any locally authenticated user with low privileges a path to full system compromise. NVIDIA GPUs power roughly 85% of the discrete GPU market and the vast majority of AI accelerator infrastructure, so CVE-2026-24192 has an unusually broad blast radius, from bare-metal Linux workstations to hypervisor hosts running vGPU Manager.
JunDong Xie of Ant Group discovered the vulnerability, and NVIDIA disclosed it in its May 2026 security bulletin alongside 13 other CVEs. It carries a CVSS v3.1 base score of 7.8 (High) and affects three production driver branches across the GeForce, RTX, Quadro, NVS, and Tesla product lines, plus the vGPU Manager and Cloud Gaming Virtual GPU Manager components.
Technical Information
Root Cause: CWE-681 Leading to Heap Overflow
The root cause of CVE-2026-24192 falls under CWE-681: Incorrect Conversion between Numeric Types. As the MITRE CWE definition states, "when converting from one data type to another, such as long to integer, data can be omitted or translated in a way that produces unexpected values." Here, a type conversion error in the NVIDIA kernel module produces an unexpected numeric value, and that value is then used to size or index a heap-allocated buffer. The result is a heap-based buffer overflow (CWE-122).
CWE-681 sits under the broader CWE-682 (Incorrect Calculation) parent and is closely related to CWE-190 (Integer Overflow or Wraparound) and CWE-680 (Integer Overflow to Buffer Overflow). The pattern shows up elsewhere in NVIDIA's codebase too. The same May 2026 bulletin includes a related CVE for an "integer overflow or wraparound" in the NVIDIA kernel module, which confirms that numeric type handling is a systemic concern across the driver stack.
CVSS Vector Decomposition
The CVSS v3.1 vector string is AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H, which breaks down as follows:
| CVSS Metric | Value | Meaning |
|---|---|---|
| Attack Vector (AV) | Local | Attacker needs local system access |
| Attack Complexity (AC) | Low | No special conditions or race conditions required |
| Privileges Required (PR) | Low | Basic user privileges are sufficient |
| User Interaction (UI) | None | No user action needed beyond the attacker's own |
| Scope (S) | Unchanged | Impact confined to the vulnerable component |
| Confidentiality (C) | High | Total information disclosure possible |
| Integrity (I) | High | Total system integrity compromise possible |
| Availability (A) | High | Total system availability loss possible |
According to cybersecurity-help.cz, the CVSS v4.0 vector is CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:U/U:Clear. In it, E:U confirms that no exploitation has been reported, and AT:N means there are no attack requirements beyond local access.
Exploitation Flow
The exploitation chain follows a well-understood pattern for CWE-681 to CWE-122 compositions:
Step 1: Type Conversion Trigger. The attacker invokes a GPU driver operation (likely an ioctl call to one of the /dev/nvidia* device nodes) that internally converts a value between numeric types, such as from a larger integer type to a smaller one, or from signed to unsigned. The driver does not validate that the conversion preserves the original value's semantics.
Step 2: Value Corruption. The incorrect conversion yields a corrupted or truncated value. Converting a large unsigned 64-bit integer to a 32-bit signed integer, for instance, may truncate the upper bits and leave a small or negative value that does not match the intended buffer offset or size.
Step 3: Heap Buffer Overflow. A later memory operation on a heap-allocated buffer uses the corrupted value. If the value is a size parameter, the allocated buffer may be too small for the data written to it. If it is an index, the write lands outside the buffer's bounds. Either way, the result is a heap overflow in kernel address space.
Step 4: Privilege Escalation. The NVIDIA display driver runs in kernel mode on Linux, so a heap overflow in the driver's address space is a direct path to kernel-level code execution. An attacker can overwrite adjacent heap metadata or function pointers to redirect execution flow and escalate from a low-privileged user to root or kernel-level access.
NVIDIA's official bulletin says successful exploitation may lead to five distinct kinds of impact: denial of service, escalation of privileges, information disclosure, data tampering, and code execution. Treat it as a full system compromise scenario, not a single-dimension risk.
High Risk Environments
The local attack vector makes this vulnerability most dangerous where untrusted users have local shell access:
- Multi-tenant servers and HPC clusters, where multiple users share GPU resources
- Container hosts with GPU passthrough, where a compromised container could interact with the host's GPU driver
- Virtualization platforms using NVIDIA vGPU, where the vGPU Manager runs on the hypervisor host
- Cloud GPU instances offered by major providers (AWS, Azure, GCP), where millions of instances may run the vulnerable driver
In multi-tenant cloud GPU environments, a compromised low-privileged container or VM could use this vulnerability to escape isolation boundaries through the shared GPU driver.
Patch Information
NVIDIA fixed CVE-2026-24192 in Security Bulletin 5821, first published on May 19, 2026, and revised through May 21, 2026 (revision 3.0). The fix ships as updated GPU display driver binaries. NVIDIA's kernel-mode driver is proprietary, so there is no source-level patch to inspect. The remediation corrects the numeric type conversion logic in the driver's kernel-mode layer so that buffer sizes are computed with the correct numeric type widths, which prevents the overflow condition.
Linux GPU Display Driver Patched Versions
| Driver Branch | Affected Versions | Patched Version |
|---|---|---|
| R595 | All prior to 595.71.05 | 595.71.05 |
| R580 | All prior to 580.159.03 | 580.159.03 |
| R535 | All prior to 535.309.01 | 535.309.01 |
The same versions apply to the GeForce, NVIDIA RTX/Quadro/NVS, and Tesla product lines on Linux.
vGPU Manager Patched Versions
| Driver Branch | vGPU Software Version | Patched Driver Version |
|---|---|---|
| R595 | 20.1 | 595.71.03 |
| R580 | 19.5 | 580.159.01 |
| R535 | 16.14 | 535.309.01 |
Cloud Gaming Virtual GPU Manager
The fix ships in the April 2026 Release with driver version 595.71.03 for Red Hat Enterprise Linux KVM and VMware vSphere.
Important Notes
The R570 driver series is End of Life and will not receive a patch for this vulnerability. Users on that branch must upgrade to R535 or a newer supported series.
In revision 2.0, NVIDIA corrected the original advisory by removing an erroneous R570 mention for Linux and replacing R590 references with R595, along with updated driver versions. Administrators working from an early copy of the bulletin should check the latest revision.
In vGPU environments, the Virtual GPU Manager must be updated on the hypervisor host as well as the guest driver. The vGPU Manager fix versions differ slightly from the guest driver versions (the R595 vGPU Manager is 595.71.03, while the guest driver is 595.71.05), so the hypervisor and guest driver updates need to be coordinated.
Standalone display driver updates are available on the NVIDIA Driver Downloads page. vGPU and Cloud Gaming updates are distributed through the NVIDIA Licensing Portal.
Detection Methods
Automated Scanning with Tenable Nessus
Tenable released two Nessus plugins on May 22, 2026 that cover CVE-2026-24192 directly.
Nessus Plugin 316514 ("NVIDIA Linux GPU Display Driver (May 2026)") is a local check (NASL file: nvidia_unix_2026_5.nasl) that runs through the Nessus Agent on Linux systems. It checks for CVE-2026-24192 along with several other CVEs from the May 2026 bulletin. The plugin relies on the application's self-reported version number and needs the knowledge base items NVIDIA_UNIX_Driver/Version and NVIDIA_UNIX_Driver/GPU_Model to be populated. It carries a CVSS v3 base score of 8.8 (High) and is tagged with IAVA 2026-A-0493.
Nessus Plugin 316512 ("NVIDIA Virtual GPU Manager Multiple Vulnerabilities (May 2026)") covers CVE-2026-24192 in vGPU Manager deployments (NASL file: nvidia_vgpu_2026_5.nasl). It is agent-based and relies on the self-reported version of the NVIDIA Virtual GPU Manager software. It matters most for virtualized GPU environments running on VMware vSphere, Red Hat Enterprise Linux KVM, XenServer, and Ubuntu. For the vulnerabilities this plugin covers, Tenable states: "No known exploits are available."
Manual Version Verification
For organizations not running Nessus, checking the version by hand is straightforward. To see the installed driver version, run:
nvidia-smi
or read:
cat /proc/driver/nvidia/version
Compare the output against the fixed versions in the NVIDIA Security Bulletin. For standalone Linux drivers, any version prior to 595.71.05 (R595), 580.159.03 (R580), or 535.309.01 (R535) is affected. For vGPU Manager installations, the fixed versions are 595.71.03 (R595), 580.159.01 (R580), and 535.309.01 (R535).
Why No Exploit Detection Signatures Exist
No exploit-specific IoCs, YARA rules, Sigma rules, or network-based IDS signatures have been published for CVE-2026-24192, and that is expected. The attack vector is local, so network-layer detection through Snort or Suricata does not apply. The underlying weakness is a numeric type conversion error that leads to a heap overflow inside the kernel-mode driver, and with no public proof of concept, there are no characteristic crash patterns, kernel log signatures, or behavioral indicators to write host-based detection rules around. No entries for this CVE were found in the SigmaHQ rule repository or Neo23x0's signature base YARA collection.
Government and DoD Environments
Both Nessus plugins reference IAVA identifier 2026-A-0493, which can be used to track compliance and remediation status through standard vulnerability management workflows.
Affected Systems and Versions
The vulnerability affects the NVIDIA Display Driver for Linux across three driver branches and multiple product lines, plus the NVIDIA vGPU software and the Cloud Gaming Virtual GPU Manager.
Linux GPU Display Driver
| Product Line | Driver Branch | Vulnerable Versions | Fixed Version |
|---|---|---|---|
| GeForce, NVIDIA RTX, Quadro, NVS, Tesla | R595 | All versions prior to 595.71.05 | 595.71.05 |
| GeForce, NVIDIA RTX, Quadro, NVS, Tesla | R580 | All versions prior to 580.159.03 | 580.159.03 |
| GeForce, NVIDIA RTX, Quadro, NVS, Tesla | R535 | All versions prior to 535.309.01 | 535.309.01 |
vGPU Manager (XenServer, VMware vSphere, RHEL KVM, Ubuntu)
| Driver Branch | Vulnerable Versions | Fixed Version |
|---|---|---|
| R595 | All versions prior to 595.58.02 | 595.71.03 |
| R580 | All versions prior to 580.126.08 | 580.159.01 |
| R535 | All versions prior to 535.288.01 | 535.309.01 |
Cloud Gaming Virtual GPU Manager (RHEL KVM, VMware vSphere)
Fixed in the April 2026 Release with driver version 595.71.03.
R535 is the long-term support branch and is used extensively in enterprise and data center environments, so the vulnerability reaches especially far into production GPU compute infrastructure. The R570 driver series is End of Life and will not receive a patch. Users on that branch must migrate to R535 or newer.
Vendor Security History
NVIDIA's security bulletins show numeric handling and input validation weaknesses recurring in the GPU display driver codebase. The May 2026 bulletin addresses 14 CVEs across Windows and Linux drivers. The July 2025 bulletin addressed 13 CVEs (CVE-2025-23276 through CVE-2025-23288), which ZeroPath described as reflecting "ongoing security challenges in complex" driver software. The January 2026 bulletin also addressed multiple vulnerabilities.
Improper index validation, integer overflow, and incorrect numeric type conversion keep appearing across these bulletins, which suggests that NVIDIA's driver code has systemic weaknesses in input validation and numeric type handling in the kernel-mode layer. The May 2026 bulletin also includes CVE-2026-24187, rated CVSS 8.8. SecurityOnline reports its CVSS v4.0 exploitation maturity as "Proof of Concept," so at least one other vulnerability in this batch may already have exploit code in circulation.
An external researcher, JunDong Xie of Ant Group, found CVE-2026-24192, not NVIDIA's internal security team. That raises questions about how effective NVIDIA's internal secure development lifecycle is for the Linux driver stack. NVIDIA has improved its disclosure processes, including moving to GitHub-based bulletins in machine-readable formats since October 2025. The vulnerability patterns still indicate that proactive code auditing for numeric type safety in kernel modules may need strengthening.
References
- CVE-2026-24192 Detail (NVD)
- CVE-2026-24192 CVE Record
- NVIDIA Security Bulletin: GPU Display Drivers, May 2026
- Tenable CVE-2026-24192
- Tenable Nessus Plugin 316514: NVIDIA Linux GPU Display Driver (May 2026)
- Tenable Nessus Plugin 316512: NVIDIA Virtual GPU Manager Multiple Vulnerabilities (May 2026)
- Tenable CVE-2026-24192 Plugins
- Feedly CVE-2026-24192 Intelligence
- SecurityOnline: NVIDIA GPU Driver Security Update CVE-2026-24187
- Cybersecurity Help: Incorrect Conversion between Numeric Types in NVIDIA
- Gaming On Linux: NVIDIA GPU Driver Security Flaws May 2026
- NVIDIA Product Security
- CWE-681: Incorrect Conversion between Numeric Types
- CWE-122: Heap based Buffer Overflow
- NVIDIA Driver Downloads
