[TheDong]

> I'd love to find out more about how we could resolve that issue.

Perhaps you could have a privacy policy which clearly states no text from user emails will be looked at by humans.

It's possible, though not easy, for a limited form of client-side encryption to be used while still offering autocomplete; e.g. if each word is encrypted client-side with a per-client secret, an encrypted "next word" could be determined and returned without the server ever knowing the specific words and sentences it operated on (other than length). There are other caveats here.

> Privacy and security are extremely high on our priority list

The lack-luster privacy policy doesn't show that at all.

[filipt]

So sorry about that - we'll make sure to improve our privacy policy to reflect that better. Thank you for pointing it out!

We're working on client-side encryption right now actually, which will further help us be better about your privacy.

The only people who can read email sentences are our researchers - and even then the sentences are not identifiable. They are obviously bound by very restrictive NDAs.

[rajeevk]

> Perhaps you could have a privacy policy which clearly states no text from user emails will be looked at by humans.

I don't think this would be practically possible. As these guys have to improve algorithm for better suggestion, someone has to look at the suggestion and the context to improve the algorithm.