As of the current macOS 27 Golden Gate beta, it seems like AI features are enabled by default with no main switch to reliably disable all of them and, most importantly, zero transparency and configuration of which requests and tasks are handled on-device and which go to the cloud.
There may be varying opinions about on-device AI, but being on-device means it does not leak your data and files anywhere, which is a huge win. However, cloud AI necessarily requires sending the data relevant to the task off your device, and is a huge privacy breach.
While one can try blocking such requests using firewalls, blocking certain hosts and IP addresses, switching individual toggles in apps and using default write commands to toggle some options not exposed in the GUI, each of these approaches is opaque and fragile.
Let's reach out to Apple through all available channels, including the Feedback Assistant app, to demand that the division between on-device and cloud AI is clearly documented, transparent and configurable.
