Integer Overflow Vulnerability in NVIDIA Triton Inference Server

CVE-2024-53880

Summary

The NVIDIA Triton Inference Server has a vulnerability in its model loading API that allows for integer overflow or wraparound errors. This occurs when an attacker loads a model file with an excessively large size, which can exceed the limits of an internal variable. If successfully exploited, this vulnerability may result in denial of service, impacting the availability of the server for legitimate users. Organizations utilizing this server should implement measures to secure their applications against such file size exploits.

References