@GossiTheDog we process tens of thousands email samples reported by users as “bad”. A tiny percentage is actual malicious samples worth the time of an analyst, most of them are one-to-one sales pitches, unsolicited contacts or offers (“I’ve noticed your profile on LinkedIN, may I send you …” and such). AI can “read” and tag such messages based on their content, so that we can surface the real attacks that we are interested in being analyzed by a human without delay. We do all the processing locally.