lab-10-confidential-ai-tdx

Lab 10: Confidential AI with Intel TDX

Demonstrates how Intel TDX protects AI workloads by encrypting VM memory at the hardware level.

What is TDX?

Intel Trust Domain Extensions (TDX) provides hardware-level memory encryption. A malicious cloud provider or hypervisor cannot read your VM's memory - they only see encrypted garbage.

Scripts

Script	Description
1_check_tdx.py	Check if TDX is enabled on the VM
2_tdx_demo.py	Main demo: Memory attack + attestation

Quick Start

# Check TDX status
python3 1_check_tdx.py

# Run the demo
python3 2_tdx_demo.py

What the Demo Shows

Part 1: Memory Attack

• Standard VM: Hypervisor extracts SSN and model weights (plaintext)
• TDX VM: Hypervisor sees encrypted garbage

Part 2: Attestation

• TDX can cryptographically prove protection to third parties
• Without TDX, you just have to trust the cloud provider

Requirements

• Standard VM: Any GCP VM (e.g., e2-standard-4)
• TDX VM: C3 machine type with --confidential-compute-type=TDX