Research / 05
Confidential AI
We explore how CPU, GPU and accelerator TEEs can support private AI workflows without obscuring their assumptions or operational limits.
Discuss this areaQuestions
What we want to understand.
Which parts of an AI stack must be trusted?
How should model and data owners verify execution?
Technical context
Architectures and systems in scope.
This list describes research scope, not endorsements, partnerships or completed evaluations.
01GPU TEEs
02Private Inference
03Model Protection
04Confidential VMs
Related directions
Work is being shaped here.
Project entries describe intended research directions and remain subject to refinement.
UniversalTEE
Portable and configurable trusted-execution research across heterogeneous systems.
Research direction
TEE Security Analysis
Security evaluation of SGX, TDX, SEV-SNP and emerging confidential-computing architectures.
Research direction
Confidential AI Infrastructure
Protecting AI models, data and inference using trusted hardware.
Research direction