> How do you know that it's actually off?

I take your meaning, in the sense that there's no inherent reason to trust one but not the other; but I think that it's fair to say that there's a big difference between:

    Hey, wouldn't it be handy for our users if we started storing and pre-processing audio *before* hearing 'Alexa', so that we're ready to respond instantly?  Let's quietly take down the text that says that we don't do that.

(which is a plausible reasoning process somewhere down the line) on the one (Echo) hand, and

    Hey, wouldn't it be a good idea if we ignored our users' explicit election to turn off a feature?

on the other (Siri) hand.