> I don’t see how Amazon employees can troubleshoot the Alexa application without both reviewing the audio files and the contextual information provided.

From my point of view, the issue isn't really that Amazon does this. The issue is that Alexa users are surprised that Amazon does this. Their surprise indicates that no real consent was obtained.

If Amazon had done something like, when a user flags an incorrect response, asking the user for permission to use that data, I would not have any issue with it.