Collection · 5 articles
Email security and privacy.
You approve an assistant's access to your mail on one screen, and that screen explains less than the decision deserves. Here's the rest of it, one question at a time: where a message goes once a model has read part of it, and how long each copy lives; what to check before you click allow, including what already has access and whether the choice is yours or your employer's; what an email written to instruct the software rather than you looks like, and what stops it; where your account keeps the record of who reached it, and which of those events can still be reversed; and which arrangements sold as encrypted email put a message past your provider. Each question stands on its own.
-
Where your mail goes when AI reads it
A message arrives, a model reads part of it, and something is kept. Here is the whole path, who else touches it along the way, and how long each piece lives.
-
What to check before you connect your inbox
The checks worth making before you click allow: what already has access, whose decision it is, and what the permission screen is actually telling you.
-
When an email is written for the AI and not for you
Some messages are aimed at the software reading your inbox rather than at you. What that attack looks like, what actually stops it, and what you still do by hand.
-
How to check your email activity, and what you can undo
Where the record of who reached your account actually lives, how far back each one goes, and which of the things that happened can still be taken back.
-
Encrypted email, and what an AI assistant cannot read
Five arrangements get called encrypted email. Only two put a message past your provider's reach, and each one costs you something once an assistant is connected.