Triton is Nvidia’s open-source inference server designed to optimize AI model deployment, now at the center of newly disclosed security vulnerabilities.