Path Traversal in NVIDIA Triton Inference Server

Executive Summary

This vulnerability in NVIDIA Triton Inference Server for Linux allows a user to perform path traversal attacks by providing a malicious path in the model name to the Triton MLflow plugin. This could enable reading, writing, or modifying files outside the intended model repository.

Useful 0 Not Useful 0

Detection Guidance

To detect this vulnerability, check if the Triton Inference Server is running with the MLflow plugin enabled. Inspect logs for path traversal attempts in model names. Verify file access patterns outside the model repository. No specific commands are provided in the resources.

Useful 0 Not Useful 0

Impact Analysis

Exploitation of this flaw may result in denial of service and information disclosure. An attacker could access sensitive files or disrupt services running on the affected system.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability could lead to information disclosure, which may result in unauthorized access to sensitive data. For GDPR, this could violate principles of data protection and privacy, potentially leading to regulatory penalties. For HIPAA, unauthorized access to protected health information could result in compliance violations and breaches of patient confidentiality.

Useful 0 Not Useful 0

Mitigation Strategies

Update Triton Inference Server to a version beyond 26.02. Disable the MLflow plugin if not required. Restrict file system permissions to limit access to sensitive directories. Monitor for unusual file access patterns.

Useful 0 Not Useful 0