Confidential Computing: Protecting Data in Use with TEEs on CPUs and GPUs
Written by AI from the linked story. Not reviewed by an editor. How this works
Traditional encryption secures data at rest and in transit, but leaves data exposed in plaintext while being processed in memory — a critical vulnerability in AI data centers handling sensitive model weights and proprietary datasets. Confidential computing addresses this gap by leveraging Trusted Execution Environments (TEEs) to isolate sensitive computations from the host infrastructure, hypervisors, and even privileged administrators. The extension of TEEs to GPUs, such as NVIDIA's Hopper architecture, is particularly significant as AI workloads increasingly rely on GPU processing for large-scale model training and inference. Without these protections, any compromise of the underlying infrastructure — whether by a malicious insider, cloud provider breach, or supply chain attack — could expose the most sensitive data at its most vulnerable moment. Organizations running AI workloads must treat data-in-use protection as a first-class security requirement, not an afterthought.
Tactical Insight
Immediate actions
- Audit all AI and data processing workloads to identify where sensitive data is processed in memory without TEE protection.
- Inventory GPU and CPU hardware to determine which systems support confidential computing capabilities (e.g., Intel TDX, AMD SEV-SNP, NVIDIA Hopper TEE).
Long-term improvements
- Migrate sensitive AI workloads to confidential computing-capable infrastructure and enforce TEE usage via infrastructure policy.
- Establish a hardware procurement standard that mandates confidential computing support for all new AI data center acquisitions.
- Implement remote attestation processes to cryptographically verify TEE integrity before processing sensitive data.
Detection & Monitoring measures
- Deploy logging and monitoring on TEE attestation events to detect tampering or unauthorized environment changes.
- Integrate confidential computing compliance checks into your CI/CD and infrastructure-as-code pipelines to prevent misconfigured deployments.