NVIDIA Announces “Security for AI” DOCA Argus Framework at RSA Conference
NVIDIA announced a new software framework that intends to provide a real-time security function for AI Infrastructure, which can be called “Security for AI.” The AI giant is using its BlueField Data Processing Unit (DPU) to read memory in GPU-based AI clusters (NVIDIA calls these “AI Factories”) and analyze it in “run-time” for security risks. We expect numerous companies will use the DOCA Argus framework to create their own in-house software systems or branded software for enterprises to use. In offering BlueField DPUs to provide cybersecurity for AI Factories, NVIDIA is creating a new cybersecurity category that CPUs have historically served. AI workloads are growing very rapidly, and this growth is driving a need for dedicated Security for AI.
We learned from NVIDIA that it has already validated Argus in its in-house AI Factories, so it has tested it against real-world scenarios. The advantages of using the Argus framework include: (a) since the Argus system resides on a DPU, it is specifically not part of the CPU processing and system and, therefore, is harder to hack, (b) using a DPU “accelerator” approach does not burden the CPUs or GPUs of the AI Factory and because it is a dedicated process, has no impact on AI processing speed, (c) using Argus, there is no need to install agent software on the host computer, (d) importantly, the Argus platform will stream data to XDR and SIEM systems so that the cybersecurity workflow can remain similar to that of existing IT workflows.
