A critical code injection vulnerability in the NVIDIA NeMo Framework allows local attackers to achieve remote code execution and data tampering.

The vulnerability exists in the model loading component, which improperly handles pathnames during the processing of .nemo files. An authenticated local attacker with low privileges can trigger this flaw by providing a maliciously crafted metadata file to the framework.

The ability to execute arbitrary code on systems running the NeMo Framework poses a severe risk to organizational data integrity and system availability. With a CVSS score of 7.8, this flaw is categorized as High severity, as it allows for unauthorized data tampering and potential lateral movement within the environment. Successful exploitation could lead to full system compromise if the framework is running with elevated privileges.

Immediate Action: Update the NVIDIA NeMo Framework to version 2.3.2 or later as specified in the vendor security advisory.

Proactive Monitoring: Monitor system logs for unusual file access patterns or unexpected processes spawned by the NeMo service.

Compensating Controls: Restrict access to the directories where model files are stored and ensure that only trusted sources can supply input files to the NeMo Framework.

Public Exploit Available: Unknown

Given the potential for remote code execution and total system impact, organizations should prioritize patching the NVIDIA NeMo Framework to version 2.3.2 immediately. Security teams should audit environments where this framework is deployed to ensure that input validation controls are strictly enforced and that no untrusted users have the ability to upload or modify model metadata files.