Confidential Computing
Confidential Computing explained simply
When we talk about data security, we usually think of two moments: data at rest (on a disk) and data in transit (on the network). Both have been encrypted for many years.
The third moment is the hardest: data in use, while the processor is working with it. At that point it usually sits unprotected in memory, and whoever controls the server (for example, a cloud provider) could, in theory, see it.
The core idea: the enclave
Confidential Computing uses special hardware to create an isolated area called an enclave or Trusted Execution Environment (TEE). Code and data inside it stay encrypted in memory, and neither the operating system nor the server administrator can get in.
In short: you trust the chip, not the people who run the machine.
Where it is used
- Training and running AI models on sensitive data (healthcare, banking).
- Sharing data between companies without either one seeing the other's data.
- Kubernetes workloads with extra isolation guarantees.
What does it cost?
Protecting data in use has a price in performance and latency. Measuring that cost rigorously is part of my research at TU Dresden, and I will share results here on the blog.