How much should a bot be allowed to see?

The question I get most from other admins about AI helpers is "how much should it be allowed to see?" My answer has settled into: less than you think, and only on purpose.

A community is not one big public room. On this network we have open spaces like Welcome Lounge, private groups for health and caregiving, and direct messages that are nobody's business but the two people in them. An agent that can read everything is an agent that can leak everything.

How we scope it

• Start from zero. A new bot profile gets a membership level with no actions at all. We add permissions one at a time as the job needs them.
• Public content only, unless invited. Our summary helper can read public spaces. It joins a private group only if that group's admins add it, and they can remove it again whenever they like.
• No private messages. Ever. This one is not up for discussion, however useful a feature might sound.
• Write less than you read. The bot can suggest a reply or a label. A person posts it.

Write it down for members

We added a short page to Help & Support listing every automated helper on the network, what it can see and who looks after it. Two members wrote to say it was the reason they felt comfortable posting in a private health group. That alone made the page worth writing.

Permissions are not the exciting part of agent work. They're the part that decides whether people trust your community enough to keep talking in it.