Practical points for anyone using a personal AI agent

Personal AI agents are useful precisely because they can reach your email, calendar and messages. That reach is also the risk. These points come from my own experience of an agent that went further than it was asked.

1. Connect only what the task needs. If it’s helping you plan a trip, it doesn’t need your inbox. Access granted for one purpose may be used for others.
2. Choose the narrowest permission available. Read-only before full access, one calendar or folder before the whole account.
3. Set the boundaries at the start, in writing. Tell it what it may and may not look at before it starts, not after it surprises you.
4. Notice what it knows. If it mentions something you never told it, ask where it came from. That is how overreach usually shows itself.
5. Treat its promises as instructions, not controls. An agent confirming that something is “off limits” is still the agent holding the record. The real control is the permission settings, which you hold.
6. Check your connected apps regularly. Your Google, Microsoft or Apple account lists which services can reach your email, calendar and contacts. Remove what you no longer use.
7. Keep sensitive material out of reach. Banking, medical, legal and family correspondence should not sit where an experimental tool can read it.
8. Remember that encrypted is not private from the agent. Encryption protects your messages from outsiders, but the agent is the other party to the chat and its operator processes what you send.
9. Judge it by its reach, not its intelligence. An agent that seems unsophisticated can still read everything you have connected.

Inside an organisation, the same risks apply with more at stake: more systems, more data, and accountability that must reach the board. That is the work of the Board AI Oversight Diagnostic.