How does AI use tools and connections?
A model on its own does not have access to everything. The app can give it a limited set of tools.
What is it?
A model on its own has no access to the internet, email or files. An application can give it tools, such as search, a calendar or a database, through connections and protocols. Each tool means a new capability and a new permission.
An everyday example
A chatbot connected to your calendar can suggest a meeting time. If it can also send invitations, it can send them in your name.
Why does it matter?
Permissions decide what AI can actually do. Permissions that are too broad increase the damage from a mistake or misuse.
A common misconception
That permission to read means only reading. Check whether the tool can also delete, send or pay.
In more detail
What is a tool call?
The model suggests using a function, and the system checks the permission and, if appropriate, runs it.
What are connection protocols?
Agreed ways for apps and services to talk to each other. Examples are APIs and protocols for connecting models to tools.
Can reading a web page be risky?
Yes. Malicious text on a page can try to slip instructions to the agent. This is called prompt injection.
What reduces the damage?
Giving only the permissions that are strictly needed, keeping data separate from instructions, logging actions and requiring approval for sensitive steps.
Remember: Permission to read is not permission to delete or send.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.