[woodruffw]

I think it's not that people don't trust Microsoft with their data (they obviously do if they use Windows, as you note), it's that they don't trust Microsoft to have performed basic diligence on data control within this feature.

[indymike]

> it's not that people don't trust Microsoft with their data

Let me fix that... People don't trust Microsoft.

[woodruffw]

Per the GP, there's a great deal of evidence that people have a revealed preference for trusting Microsoft, if they're already in the Microsoft ecosystem.

(I don't particularly trust Microsoft, but I'm also not in their ecosystem.)

[indymike]

This assumes those people have a credible alternative, which in most cases they do not.

[woodruffw]

25 years ago (or even 15 years ago), I would have probably agreed. But I don't think that's true anymore; to my understanding, students in public schools are equally as likely to use Chromebooks as they are to use Windows or Mac machines.

[indymike]

Yes, and that decision is not made by the user.

[woodruffw]

Yes, the point was that the "you don't have a choice but to use Windows" market is smaller than ever, meaning that the slice of the pie that's "I want to use Windows or don't care what I use" is larger.

[Spivak]

It's on the user side of the airtight hatchway, that's good enough. If you could read my browser history you could read this and vice versa.

[woodruffw]

If I remember correctly, one of the original concerns was that the feature didn't respect user boundaries on the host machine itself. In other words: multiple users sharing a machine could inadvertently (or intentionally) retrieve information about each other.

That would be a straightforward example of "can't read the browser history for a user, but could read it indirectly via the agent."