A back-of-the-envelope calculation says that it would cost significantly less than the annual budget of the (London UK) Metropolitan police to put every window in London under laser-microphone surveillance.

Likewise, I assume that every sufficiently important person is targeted with some form of surveillance by at least one organised criminal gang interested in either taking their money or corruption, both of which can be achieved by surveillance-enabled blackmail. “Sufficiently” is variable depending on much this tech costs, and it’s constantly getting cheaper.

How does it work without being surveilled? Most common AI use cases require data. A voice assistant needs to know my voice to work, a health assistant would need health data (weight, heart rate, etc), a scheduling assistant would need my schedule, etc etc. Or perhaps I’m misunderstanding your point?

The voice assistant needs to know your voice, but it doesn't need to share it with either a corporation or a government.

Likewise, the health assistant needs your health data, but doesn't need to share it.

The data should be stored on hardware you control, and not shared with anyone else.

It's the difference between using a computer for information keeping and other people using that information for unauthorized and malicious ends.

Which is also what makes surveillance inevitable, as long as everyone is using Someone Else's Computer (the cloud).