You may be underestimating the costs of _not_ archiving large amounts of potentially-sensitive data in a way that is accessible to a large number of employees (

You may be underestimating the costs of _not_ archiving large amounts of potentially-sensitive data in a way that is accessible to a large number of employees (and therefor not very secure). Alexa needs training. Training requires data, as real as possible and as much as possible. It also requires reviewers, and that review process requires a lot of person-hours. So if Amazon were to decide not to store voice recordings of real-world interactions (or potential-interactions like "my friend Alexa said..."), or if they were to decide not to make that data available broadly (to thousands of employees), then they would also be wiping out millions of dollars of market cap. The way I see it, it's not really a question of whether or not they're being cavalier with private data. For me that threshold was crossed with the large human voice-to-text farms. I don't want the list of people who might hear my private conversations that maybe contained syllables sounding like "Alexa" or "Siri" or "Google" to be thousands strong. That's orders of magnitude more than could ever be reasonably expected to constitute a secure circle.