Executive brief
NVIDIA Triton Inference Server is a software platform that enables high-performance AI model serving. A path traversal vulnerability in the MLflow plugin allows attackers to read, write, or modify files outside the intended model repository, potentially exposing sensitive data or disrupting service availability.
Technical details
The vulnerability is a path traversal flaw in the Triton MLflow plugin that fails to properly sanitize model names provided by users. An attacker can supply a crafted path (e.g., containing "../" sequences) as a model name to access files outside the configured model repository directory. No authentication is required—the plugin accepts user-controlled input directly. Successful exploitation allows reading arbitrary files (information disclosure), writing or modifying files outside the repository, and potentially causing denial of service. A patch is expected from NVIDIA.
Affected products
- NVIDIA Triton Inference Server <UNKNOWN>
Timeline
- 2026-08-04: disclosed