NVIDIA Container Toolkit Vulnerable to Code Execution Attacks

NVIDIA has issued a critical security update to address a high-severity vulnerability discovered in the NVIDIA® Container Toolkit for Linux.

The flaw, tracked as CVE-2025-23359, could allow attackers to exploit a time-of-check time-of-use (TOCTOU) vulnerability to gain unauthorized access to the host file system.

This vulnerability could potentially lead to code execution, denial of service, privilege escalation, data tampering, and information disclosure.

Details of the Vulnerability

The vulnerability, affecting all versions of the NVIDIA Container Toolkit up to 1.17.3 and NVIDIA GPU Operator up to 24.9.1, has been assigned a CVSS v3.1 base score of 8.3 (High).

Using a specially crafted container image, attackers can exploit this flaw to manipulate the container runtime environment, compromising the host system.

Impact

• CWE-367: The vulnerability stems from an insecure handling structure in files, allowing attackers to bypass default safeguards.
• Potential Consequences: Code execution, system crashes, elevated privileges, data exposure, and tampering.

NVIDIA acknowledges that the true risk depends on specific configurations, emphasizing the importance of evaluating your systems individually.

Security Patches and Mitigation

NVIDIA has released fixed versions of its software to remediate the issue:

• Container Toolkit: Updated to version 1.17.4
• GPU Operator: Updated to version 24.9.2

Users are advised to download and install the updated versions as per the instructions in the official NVIDIA documentation. Failure to update may leave systems vulnerable to exploitation.

The patch alters the default behavior of the NVIDIA Container Toolkit. CUDA compatibility libraries within containers are no longer automatically mounted to the default library path.

While this change enhances security, it may disrupt some applications reliant on previous behavior.

For users requiring the legacy setup, NVIDIA has introduced a feature flag, allow-cuda-compat-libs-from-container, to opt-in.

However, NVIDIA strongly advises against using this flag as it reintroduces the vulnerability.

For applications dependent on CUDA forward compatibility, users can manually configure the LD_LIBRARY_PATH to include /usr/local/cuda/compat.

This workaround, however, may pose portability challenges when using different driver versions.

This vulnerability was disclosed by Andres Riancho, Ronen Shustin, and Shir Tamari from Wiz Research, along with Lei Wang, who independently reported it.

NVIDIA encourages users to stay informed by subscribing to their Product Security bulletins for the latest updates and guidance.

With threats on the rise, ensuring prompt updates and proactive security management is crucial to maintaining system integrity.

Investigate Real-World Malicious Links & Phishing Attacks With Threat Intelligence Lookup - Try for Free