NVIDIA has released critical security updates addressing two significant vulnerabilities in its Container Toolkit and GPU Operator that could allow attackers to execute arbitrary code with elevated privileges.
The vulnerabilities, discovered in July 2025, affect all versions of the Container Toolkit up to 1.17.7 and GPU Operator up to 25.3.0, prompting immediate security patches and mitigation recommendations from the graphics giant.
Critical Security Flaws Identified
The most severe vulnerability, tracked as CVE-2025-23266, carries a critical CVSS score of 9.0 and affects initialization hooks within the Container Toolkit across all platforms.
This flaw enables attackers to execute arbitrary code with elevated permissions, potentially leading to complete system compromise through privilege escalation, data tampering, information disclosure, and denial of service attacks.
The vulnerability exploits weaknesses in container initialization processes, making it particularly dangerous for containerized environments.
A second vulnerability, CVE-2025-23267, receives a high severity rating with a CVSS score of 8.5. This flaw affects the update-ldcache hook and allows attackers to perform link following attacks using specially crafted container images.
While less severe than the first vulnerability, it still poses significant risks including data tampering and denial of service attacks.
| CVE ID | Description | CVSS Score | Severity | CWE | Impact |
| CVE-2025-23266 | Arbitrary code execution in container initialization hooks | 9.0 | Critical | CWE-426 | Privilege escalation, data tampering, information disclosure, denial of service |
| CVE-2025-23267 | Link following vulnerability in update-ldcache hook | 8.5 | High | CWE-59 | Data tampering, denial of service |
Affected Products and Updates
Both vulnerabilities impact NVIDIA Container Toolkit installations across all platforms and the GPU Operator on Linux systems.
Organizations running Container Toolkit versions up to 1.17.7 or GPU Operator versions up to 25.3.0 are vulnerable and should immediately upgrade to the patched versions: Container Toolkit 1.17.8 and GPU Operator 25.3.1.
While permanent fixes require updating to the latest versions, NVIDIA provides temporary mitigations for organizations unable to immediately upgrade.
Users can disable the enable-cuda-compat hook by modifying configuration files or environment variables. For Container Runtime users, this involves editing the config.toml file to set the disable-cuda-compat-lib-hook feature flag to true.
GPU Operator users can achieve similar protection by adding the disable-cuda-compat-lib-hook flag to environment variables during Helm installation or upgrades.
Security experts recommend immediate implementation of available patches, given the critical nature of these vulnerabilities and their potential for exploitation in containerized environments.
Organizations should prioritize updating their NVIDIA Container Toolkit and GPU Operator installations while implementing the suggested mitigations as interim protection measures.
Get Free Ultimate SOC Requirements Checklist Before you build, buy, or switch your SOC for 2025 - Download Now
