[Bug] Remote Control auto-enables without user consent on new sessions

Bug Description
Remote Control auto-enables without user action.

Every new session starts with it active. I have never run /rc, --rc, or
claude remote-control. Ruled out: aliases, shell config, settings.json,
env vars, plugins. disableClaudeAiConnectors is true and it activates anyway.
~/.claude.json contains seenNotifications["remote-control-auto-on"], so this
looks intentional.

Impact: the session becomes remotely reachable, and per the docs, tool results
cross that bridge — including the contents of every file the agent reads. A
user who did not opt in has no reason to know that. On client code, that is a
disclosure nobody consented to.

The docs list three explicit activation paths and describe no automatic
behaviour.

Asking for: confirmation of whether auto-on is intended, documentation of it,
and a settings key to disable it globally. Per-session disconnect is not a
control.

Environment Info

• Platform: darwin
• Terminal: xterm-256color
• Version: 2.1.250
• Feedback ID: 27b0db8d-e05e-4f28-84c5-dda4a95f1af7

Errors

[]