Shared vs isolated computers

The isolation question deserves its own lesson, because it quietly changes how you work.

With Muse and Dots, each agent gets its own cloud computer. Spin up a second dot, or a second task, and it's a clean, separate sandbox: a login or file in one doesn't automatically expose it to another. Compartmentalizing is the default.

With Grok Bot, it's the opposite: every Bot you own shares one cloud computer. The files, browser sessions, and command-line logins on that machine are reachable across your whole Bot roster. xAI's docs are explicit — do not use separate Bots as a security boundary. People's instinct is "I'll put the risky task on a separate Bot to wall it off." On Grok Bot, that wall doesn't exist.

So where does safety come from when the computer is shared? Least privilege, not separation:

• Connect only the apps a task actually needs — every connection is another door into that shared machine.
• Prefer a scoped or read-only account over your full-access personal one, so even a misbehaving agent can't reach past its job.
• Sign out of a service when a task no longer needs it, and remove sensitive files — on a shared, persistent computer, whatever you leave behind stays reachable.

Even on the isolated agents (Muse, Dots), least privilege is still the right habit — it just isn't your only line of defense there. A useful rule across all three: grant the narrowest access that lets the job succeed, and nothing more. An agent can't misuse an app you never connected.

One reassuring detail on Muse: it never sees your passwords or payment methods at all — credentials go into secure storage it can use but not read, and it can't see passwords you type into the browser yourself. That's the kind of built-in limit worth looking for in any agent you trust with real accounts.