The Missing Piece
Think of data like a physical document. You can lock it in a safe (data at rest), or seal it in an envelope while mailing it (data in transit). But what happens when someone actually opens the envelope and reads it to act on it? That moment — when the document is out in the open, being read — is the gap that confidential computing addresses.
For decades, security engineering has focused on two states of data:
- Data at Rest: Encrypting files on a hard drive (AES-256). If someone steals the disk, they can't read it.
- Data in Transit: Encrypting data moving over the network (TLS/SSL). If someone taps the wire, they can't read it.
But there is a third state: Data in Use.
To actually process data — to sum numbers in a spreadsheet, train an AI model, or query a database — the CPU must work with that data in the system's RAM (main memory). Encryption cannot be applied here in the traditional sense, because the CPU needs plaintext to compute on it. This is the gap.
The Vulnerability
At this exact moment of processing, the data is exposed to several real threat vectors:
- Malicious Admin: A cloud provider administrator with physical access to the server could attach a debugger or dump the contents of RAM to a file.
- Compromised OS: If an attacker gains root access to the Operating System, they can read the memory of any running process — including your application's secrets.
- Hypervisor Attacks: In the cloud, your Virtual Machine (VM) runs on top of a Hypervisor — software that manages multiple VMs on one physical host. Theoretically, the Hypervisor can inspect any VM's memory.
- Cold Boot Attacks: RAM retains its contents for seconds to minutes after power loss. An attacker with physical access can freeze the RAM chips and read them in another machine.
Confidential Computing solves this by encrypting data while it is being processed by the CPU — using hardware mechanisms that no software, not even the operating system, can bypass.