NVIDIA BlueField DPUs Vulnerable to Critical Code Execution Exploit
A critical vulnerability in NVIDIA's BlueField DPUs and ConnectX platforms, CVE-2026-65094, allows low-privilege VM users to execute arbitrary code.
NVIDIA has disclosed a critical vulnerability, CVE-2026-65094, affecting its widely deployed BlueField DPUs and ConnectX networking platforms. This flaw, residing within the VIRTIO-Net component, carries a CVSS v3.1 score of 9.0, signaling a severe risk to data center and cloud environments.
The vulnerability is characterized as a CWE-123 write-what-where flaw, enabling a malicious actor to manipulate system memory by writing arbitrary data to unintended locations. This capability can be leveraged to achieve attacker-controlled code execution, posing a significant threat to the confidentiality and integrity of affected systems. The attack vector is particularly concerning as it can be initiated by a low-privilege user within a virtual machine (VM), making it a potent threat in multi-tenant or shared infrastructure environments.
Exploitation of CVE-2026-65094 can occur without user interaction and involves a scope change, meaning an attacker could potentially affect resources beyond the initially compromised component. This amplifies the impact in production environments where DPUs are crucial for offloading networking, storage, and security tasks. A successful compromise at this level could allow attackers to bypass existing security controls, move laterally across workloads, or disrupt critical network traffic processing, undermining the security posture of high-performance computing and cloud-native architectures.
NVIDIA's security bulletin indicates that the vulnerability impacts multiple versions of its VIRTIO-Net implementation, including both general availability (GA) and long-term support (LTS) releases. Specifically, versions prior to 25.10.6 for VIRTIO-Net GA, 25.10.2 for LTS25, 24.10.50 for LTS24, and 23.10.23 for LTS23 are affected. The company has made patched versions available and strongly urges customers to update their systems immediately to mitigate the risk of exploitation.
While NVIDIA's internal risk assessment provides an average across various deployments, organizations are advised to conduct their own evaluations to understand their specific exposure, especially in environments where untrusted virtual machines or tenants might interact with shared networking resources. The vulnerability was discovered internally by NVIDIA, and as of the disclosure, no active exploitation in the wild has been reported.
However, given the critical nature of write-what-where vulnerabilities and their historical exploitation, security teams should prioritize patching. Organizations can obtain the necessary updates through NVIDIA's official product security portal and DOCA VIRTIO-Net distribution channels. Implementing a layered defense strategy, including applying patches, restricting access for untrusted VMs, and monitoring for anomalous network behavior, is recommended to bolster security against potential attacks.
This vulnerability underscores the increasing complexity and criticality of hardware-level security in modern computing infrastructure. As DPUs become more integral to data center operations, securing these components against sophisticated attacks is paramount for maintaining the integrity and performance of cloud services and high-performance computing environments.