Okay, so it sends data after you've said the trigger phrase, but not before, that much we've all agreed on. Has anyone determined what it's sending? Hear me out here...<p>For the trigger to work, which doesn't require key presses or other physical input, the microphone must be active and keeping a buffer of live speech, right? Okay, let's step aside for a moment to examine another piece of technology that keeps a live buffer of streaming media...<p>I have a TiVo. I love that thing. One of the coolest features is this: TiVo keeps a buffer of live shows on each of its tuners so that, if you opt to record a show that's been on for ten minutes, it can save it from the beginning, literally "capturing and digitally storing the past", once I press "record".<p>Back to Chrome/Chromium: Until someone determines what it's sending, and their black box makes it difficult to see the source of what audio is encoded and how it's encoded, and given that the technology to keep a rolling buffer of 30 minutes of streaming audio and video has been around in TiVo since, well, the first TiVo... Is it really so hard to believe that Google could be sending packets containing audio spoken in the room before the trigger phrase was captured?<p>It wouldn't take much for Google to assuage this privacy concern.<p>1) Opt-in, not opt out<p>And<p>2) Show us what's in the black box (or at least publish tcpdump-verifiable specs on what you're transmitting). If there was a debug option to transmit in clear text, and a statement of the audio codec used, that would give me some sense of understanding, even though that still leaves steganography as an option. I don't have enough tin foil for that hat right now.