Confidential computing
Running code in a sealed hardware environment so that even the machine's owner cannot see the data inside.
Normally, whoever runs the computer can read what is on it. Confidential computing breaks that assumption. The processor creates a sealed area, sometimes called a trusted execution environment, where data stays encrypted even while it is being worked on, and the hardware can produce cryptographic proof of exactly which code is running inside. The cloud provider hosting the machine cannot look in.
For AI this solves a specific standoff. When an outside organisation wants to test a company’s model, someone normally has to hand something over: either the test questions, which lets the company optimise for them, or the model weights, which risks the company’s most valuable asset. Google’s Confidential Space was used in August 2026 to run a double blind evaluation of a Gemini model where neither side saw the other’s material, with the guarantee coming from hardware rather than a contract.