Well, presumably neither does the Windows version they'll ship, according to them. But it IS supposed to record all activity on the background. I don't know that it's doing OCR, but presumably that's because they don't have to, since they control more of the ecosystem. From how it was described it's closer to Timeline, where it's logging all activities you take on the device and then uses an AI search to parse them and identify them. If it can find a picture you were watching on a specific time range by its content it must be logging with enough detail to know what's on the screen at all times. And unlike MS they are not committing to not sending the data over to server, although they are comitting to not storing it.

But hey, this is a good case of them hiding just what they are pulling helping them make it sound less creepy even though it really isn't. This is the kind of reasoning I was wondering about. As I said above, it'll be interesting to see how sticky this framing is.