Anthropic has put Claude, its AI assistant, inside Google's office apps. Claude for Google Workspace, released on Oct. 6 as a public beta, an early version open to all paying users that may still change, opens in a sidebar next to a document, spreadsheet or slide deck, reads the file and makes changes in it directly. Anthropic says that saves copying text back and forth between a chat window and the file. Google already builds its own AI assistant, Gemini, into these apps; the add-on gives people who prefer Claude a way to use it there instead.
Anthropic says that in Docs, Claude can fix a sentence or restyle a heading without disturbing the formatting around it, and can offer bigger rewrites as proposals in the sidebar, which the user applies or dismisses. In Sheets it can write formulas, build charts and add new tabs. In Slides it can build new slides from the deck's own layouts and themes, then check its work and flag anything that overlaps or runs off the slide, and text that is hard to read.
The add-on is open to people on Claude's paid plans (Pro, Max, Team and Enterprise) and installs from Google's Workspace Marketplace, its store for add-ons to these apps. One install covers all three apps. Anthropic notes that many organizations block such installs by default, so at those companies an administrator has to allow Claude first. Anthropic also released connectors, in beta too, that let Claude create and edit Google files from a conversation in the Claude app. On Team and Enterprise plans, an account owner has to switch them on first.
By default, Claude shows each change and waits for the user to approve it. A second mode lets it work through a task without stopping, though it still asks first before adding a link or a web image or removing someone's access. Google records its edits under the user's name in the file's version history, so collaborators see them as the user's own edits, and they can be undone like any other change. Anthropic says the add-on sees only the file it is opened in, plus any other services the user has linked to Claude, that it cannot share the file or change who has access, and that it does not use this content to train its models.
As a beta, it has gaps. It does not work in Firefox, cannot read or reply to comments in Docs, and stops if the sidebar is closed. Anthropic also warns that a file can carry instructions planted by someone else, sometimes hidden in white or tiny text, meant to steer Claude. It says Claude treats a document's contents as information, not commands, and is designed not to obey a file that tells it to share the file, open a link or contact anyone, but it still advises people to review its proposed actions, especially in files from outside their organization. The company says not to use the add-on with protected health information, the patient records that U.S. health privacy law guards, and it has not said when the beta will end.