SpaceXAI's Grok now works inside Google's document editors. A free add-on opens a side panel next to a Doc, Sheet or Slides file, where the chatbot can turn notes into a formatted draft, build a deck from an outline, or answer questions about a spreadsheet and point to the cells it used. One install covers all three editors.
Free is what makes it spread
Google's own assistant, Gemini, comes with paid Workspace plans, and personal accounts need a paid Google AI plan. Claude has a similar side panel, but it needs a paid Claude plan and is still in beta. Grok costs nothing. So the person at a company that pays for no AI tool can add one to the files they already work in, without asking anyone.
That is how a tool usually enters a workplace: one employee finds it useful and the rest copy them. It has happened with spreadsheets, file sharing and chat apps, and there is no reason AI panels will be different. Google has been linking outside apps into Gemini. Grok goes the other way, putting an outside AI into Google's own files.
What the screen hands over
The permission screen is more specific than most people read. Grok asks to see, edit, create and delete all of an account's Docs and all of its Slides presentations, even when it is installed from a single file. For Sheets it asks only about spreadsheets where it has been installed, and for Drive only about files used with it. It also asks to connect to an outside service, which is how what it reads gets sent to SpaceXAI.
The edits are not the worry. Changes in Docs show up in version history, and changes in Sheets are ordinary edits that can be undone. The worry is where the text goes.
Picture a bookkeeper at a 30-person accounting firm. Her meeting notes in a Doc include a client's payroll figures, and she asks Grok to tidy them into a clean summary. The summary is good and she has done nothing unreasonable. But the notes have left the firm, nobody there decided that they could, and what SpaceXAI does with them is set by its privacy policy, which she did not read. Multiply that by every employee who finds a free panel useful, and by every AI vendor that will offer one, and a company ends up with several outside AI tools reading its files without ever having chosen any of them.
The screen alone is a weak safeguard, because people approve what looks like part of the tool they are already in. In 2017 a fake app named "Google Docs" got about a million Gmail users to approve access to their email in about an hour, using Google's own approval page. That is not a flaw in Grok, which is a legitimate product. It is a fact about human habit that every AI vendor now benefits from.
The control is one setting
Google lets an administrator choose whether staff can install any Marketplace app, only approved ones, or none. An administrator can still install an app for everyone. So the workable policy is neither a ban nor a free-for-all. It is to allow only approved apps, read the permission screen of each AI panel once, and install the ones the company trusts for the people who need them. That takes about an hour, and it moves the decision from each employee's click to a person who has read what the panel asks for.
For an individual, the step is smaller. Before pressing Continue on any AI add-on, read which files it can reach and whether it sends them elsewhere, and try it first on a document with nothing confidential in it.