Breaking NVIDIA Triton: CVE-2025-23319 - A Vulnerability Chain Leading to AI Server Takeover

Wiz Research disclosed a chain of critical vulnerabilities in NVIDIA Triton Inference Server's Python backend that allow an attacker to obtain the backend's internal shared memory name via an error message, register that internal shared memory through the public shared-memory API to gain arbitrary read/write into backend memory, and ultimately achieve remote code execution and full server compromise; NVIDIA assigned CVE-2025-23319, CVE-2025-23320, and CVE-2025-23334 and released patches, with users advised to update to version 25.07.