What breaks if a Dot and a Grok Bot are both signed into the same email?

Nothing in either product breaks. The mailbox does. A Dot and a Grok Bot signed into the same email are two clients that do not share a lock, a draft folder, or an approval. OpenAI’s getting-started page says you connect your personal email, and that at launch a Dot cannot have its own standalone address. Grok Bot’s plugin help says the Gmail connector can search and read, draft and send, and apply labels, on one mailbox at a time, and that an installed plugin is available to every Bot on the account. If both are allowed to send, both can send. If either archives or labels, the other sees the mailbox that action left behind.

This is not a documented bug with a single error message. It is two cloud computers, plus every other Grok Bot on that account, treating one inbox as theirs. The vendors do not coordinate. You do.

Which login did you actually give each one?

There are two doors on the Grok side, and they are not the same door. The Gmail plugin is a connector. Cursor’s docs say plugin logins stay off the computer: the OAuth token sits on the connector backend, and the Bot calls tools without receiving that token. The browser on the cloud computer is the other door. Browser cookies and signed-in sessions are shared by every Bot on the account. Signing into Gmail in that browser makes the session available to the other Bots. Disconnecting the plugin does not sign the browser out. The approvals page says to sign out of a service on the shared computer when it should no longer be available, and to delete the connector and revoke it in the source service as well.

A Dot has its own cloud computer, with its own browser sessions. It does not see the Grok Bot’s cookies. Connecting email for a Dot is a ChatGPT plugin on your account. The privacy FAQ says those plugin permissions apply across Dots, ChatGPT, ChatGPT Work, and Codex. A mailbox you connected so the Dot could read it is a mailbox ChatGPT can use within those same permissions. The getting-started page also says the Dot can review connected information on its own and form memories from it, even when you did not ask a question about that mail.

So “both signed in” can mean four things at once: the Gmail plugin, the Grok browser session, the Dot’s email plugin, and a login sitting in the Dot’s cloud browser. Any one of them can read. Any of them that includes send can send. I have not found a page on either site that says these sessions know about each other.

What can the two of them do to one thread?

They can each read it, and they can each change it if you gave them change. Gmail’s unread state, labels, and archive bit live on the mailbox. Grok’s connector is documented as able to apply labels. A Dot whose app permission allows changes can change records too. ChatGPT’s app-permission article lists the settings a personal account may see after a plugin is installed: Always ask, Allow read actions, Allow low-risk actions, and Allow all actions. Allow read actions means ChatGPT can read without asking and asks before making changes. Sending an email is one of the examples that article puts on the change side, along with editing or deleting a record. Allow all actions means supported actions can run without another prompt, and the article calls that elevated risk. Not every app shows every setting. If your mail connection does not offer Allow read actions, do not invent it. Use the strictest choice the screen actually shows, or do not connect it.

There is no queue between the products. Nothing in the docs says a Dot waits while a Grok Bot holds a thread, or the reverse. Two replies can land on one conversation because two agents were both told, in their own chats, to answer the mail. Two labels can land because both were told to file it. I am not going to describe a trick that forces that collision. The ordinary case is enough: you asked each one to “handle the inbox,” and the inbox has one copy of each message.

Drafts are a smaller version of the same split. Grok’s connector can draft and send. On the Dot side, the learn docs say that asking it to draft replies does not give it permission to send them. A draft is still a message in a mailbox both can see, if both can read drafts. One can send a draft the other wrote, if the one that sends is allowed to send. That is not a clever failure. It is send permission plus a shared drafts list.

Who else on the Grok account can see that mailbox?

Every Bot you run. The plugin help is direct: an installed plugin is available to every Bot on that Grok Bot account, and you disconnect Gmail if a Bot should not use that mailbox. You cannot fence one Bot off while leaving the connector installed. Gmail also connects one mailbox at a time. A personal mailbox and a work mailbox cannot both be connected to that Gmail plugin. To switch, you disconnect one and authorize the other.

The browser path is wider than the plugin. The Bots share one cloud computer and its logins. A webmail session you opened so one Bot could “just look” is a session the others can open. Screens are separate work surfaces, not security boundaries. One Bot runs one computer-use task on its screen at a time. A different Bot can still be in the same mailbox through the plugin while the first is clicking.

A teammate’s Team Bot is a different ask. Cursor’s docs say a teammate’s Team Bot asks before it uses your accounts, and plugin logins are not the same thing as the shared computer between your own Bots. Do not read “my Bots share a mailbox” as “my whole company does.” Between users, the computer is isolated. The collision in this article is yours: your Dot and your Grok Bots, on a mailbox you connected twice.

What does not get undone when you disconnect?

Mail already sent stays sent. Labels already applied stay applied until someone changes them in the mailbox. A Dot that formed memories from the connected account keeps what it learned until you delete that memory or reset the Dot. Disconnecting an app stops future access. The privacy notes on Dots say it does not delete information the Dot already obtained, and that deleting the Dot is what deletes the Dot’s conversations, memories, and scheduled tasks. It still does not unsend a message or undo a change in the mail provider.

On the Grok side, removing the connector is not the whole job. Revoke the app in the Google account as well, and sign out of Gmail in the shared browser. A routine that fires on new mail will keep firing until you pause or delete it. The mobile docs say you can pause a routine from the Bot’s profile. Editing the schedule is a desktop step. A routine you forgot is a third client of the same mailbox, running while you are comparing the Dot and the Bot in a chat.

Proactive research on the Dot is the quiet reader. The announcement says that when you are not actively working with it, the Dot looks for ways to help using apps you already connected, with tools restricted to read-only. Those tools cannot send messages, change app content, or control the browser or computer. Any follow-up action is under the normal permissions. Read-only research can still mean the contents of the mailbox have been pulled into the Dot’s context and memories. “It cannot send” is not “it never saw the mail.”

What should stay human if both products are in your life?

Send, and the mailbox you cannot afford to have filed twice. Pick one product as the reader if you want a reader at all. Give that one the connection. Leave the other signed out, plugin removed, browser session gone. If the reader is the Dot, choose Allow read actions when that setting is actually on the screen, so a read does not become a send. Add a custom rule that says not to send mail if you want a standing instruction, and remember the FAQ: a custom rule cannot override built-in safety, and the Dot can still make mistakes. Auto-review, in that FAQ, checks the recipient and the message before a send. Your approval cannot override a block. None of that coordinates with Grok Bot. Grok’s own card is Allow once, Always allow, or Deny for the action it proposed. Always allow on “send mail” is how a second send happens on a later thread you did not look at.

Do not sign the same mailbox into both cloud browsers “so they can see the thread.” The plugin is the narrower door, and even the plugin is account-wide on the Grok side. The browser is the wide door. A password typed into either chat is the wrong door entirely. Take over the screen if a login is required, on the product that is supposed to have the mailbox, and do not repeat it on the other.

If both are already connected, the repair is subtraction, not a smarter prompt. Disconnect one. Revoke it at the provider. Sign the browser out. Pause the routine. Then look at the mailbox itself for two replies, a label you did not mean, or a draft that went out. The agents will not reconcile that for you, because neither one was built to know the other was there.