[bell-cot]

Ooh...

Admitting that I've done zero checking of this "agent"'s security - any AI that's replying to people's emails, at scale, is just begging for malicious parties to try to game - to make it help them perform password resets and all sorts of other lucrative mischief.

[nathanganser]

Totally! that's why it's only drafting emails and not actually sending them out.

A complete autopilot support agent would be cool but there is to much risk in that :)