Executive brief
NVIDIA Triton Inference Server is a software platform that allows organizations to deploy machine learning models for inference at scale. This vulnerability allows an attacker to bypass path restrictions and access files outside intended directories, potentially leading to unauthorized data access or remote code execution on servers running the affected software.
Technical details
The vulnerability is a path traversal flaw in NVIDIA Triton Inference Server for Linux that allows an attacker to bypass path validation controls and access files outside the intended directory structure. The root cause is insufficient input validation on file paths, enabling absolute path traversal attacks. An attacker with local access to the server can exploit this by specifying absolute paths in API requests or configuration parameters to read arbitrary files or potentially write malicious content. The vulnerability could lead to information disclosure (reading sensitive model files, configuration data, or system files) or code execution if combined with other attacks. Patches are available from NVIDIA in security bulletin 5865 released in August 2026.
Affected products
- NVIDIA Triton Inference Server <UNKNOWN>
Timeline
- 2026-08-18: disclosed