Security

Android's "AI seal": a locked room for on-device AI

Google says a new hardware-isolated vault will keep the data your phone's AI works on separate from the rest of the system. What that means in plain words.

By Yemini Team, 2026-11-13

When an AI assistant on your phone summarises your messages or plans your day, it has to read personal data. A fair worry is what happens to that data if something else on the phone is compromised. In October, Google announced a design it calls AI seal, and Privacy Guides covered it. The idea is a protected vault on the phone where AI work happens, kept apart from the rest of the operating system. How it works, simply The vault runs in a protected virtual machine, built on Android's Virtualization Framework and the pKVM hypervisor. The goal is that your personal data stays isolated even if the main Android system were fully compromised. Several AI services can share one vault. The vault does not hand out raw data. If an assistant summarises your schedule, only the summary leaves the vault. Who gets it Google says phones with MediaTek's Dimensity 9600 Pro and Qualcomm Snapdragon chips are expected to support the design, and it is working with phone makers. Google also describes a longer-term goal of running the whole on-device AI inside such a vault, using an open standard. What to take from it This is a promise about design, not a guarantee, and it arrives with new phones rather than as an update to old ones. It is still a healthy direction: do the sensitive work on the device, and keep it walled off. That is the same principle we follow in Yemini apps, where your notes and files stay on your phone unless you choose to share them.