Nvidia Triton Vulnerabilities Pose Big Risk to AI Models

Wiz has uncovered multiple vulnerabilities in Nvidia’s Triton Inference Server, which could allow remote attackers to take control of servers and compromise AI data. These findings highlight ongoing risks to AI systems relying on Nvidia products, especially in cloud environments. #NvidiaTriton #AIModelSecurity

Keypoints

Wiz discovered three vulnerabilities in Nvidia’s Triton Inference Server, affecting AI deployment.
Exploiting these flaws can lead to remote code execution, data theft, or server control.
CVE-2025-23319 and CVE-2025-23320 pose high-severity risks impacting the Python backend.
The vulnerabilities can be chained to escalate from minor leaks to full system compromise.
Organizations using Triton should prioritize patching to mitigate potential breaches.