Sovereign AI & Confidential Computing in Action: Red Hat, Intel & NVIDIA

Intel’s Principal Engineer Phil King demonstrates how confidential computing can protect sensitive AI models and data from memory-based attacks. Using Intel Trust Domain Extensions (Intel TDX) with NVIDIA GPUs, the demo first exposes how an unprotected VM can leak a user’s query and the AI model’s response through a memory dump, then demonstrates how those same data remain inaccessible once confidential computing is enabled. Red Hat AI provides the unified control plane and uses confidential containers to isolate AI workloads in secure “vaults,” helping protect sensitive data across the CPU and GPU environment.