The main thing compared to the old voice mode is that your speech instead of being fed to a speech to text model is now fed to the LLM directly. I'm not sure if

The main thing compared to the old voice mode is that your speech instead of being fed to a speech to text model is now fed to the LLM directly. I'm not sure if that would be enough to trigger a need for external review though. If they were using your voice recording for training purposes, then of course absolutely. Which then for a good reason. Or it could also be about the way the voice recording is transported or where it's stored, for how long it's stored etc. Because speech to text could be achieved even offline and not in cloud and be there for shorter period of time. You can delete the recording and just use the text. If the voice recording is tokenized in an ongoing conversation then it would be stored indefinitely.