As much as I can recall, they said most of the commands would be processed on-device and only a few would be send over to the cloud with the user’s permission. But I get the point: Cloud can be scary and send info to ChatGPT sounds even more scarier knowing the brains behind it and how they operate. I would love Apple to post a deep-dive into the system architecture behind this just like they did with iMessage PQ3 protocol a few months ago.