◆ AI AT WORK · Tools
Claude's edits from inside Google Docs land under your name
The News
On October 6, 2026, Anthropic made Claude for Google Workspace a public beta on all paid Claude plans. The add-on opens Claude in a sidebar inside Google Docs, Sheets and Slides, where it reads the open file, including any text, cells or slides the user has selected, and edits it in place. In the default "Ask before edits" mode each change waits for the user's approval; an "Accept all edits" mode applies ordinary content edits without stopping. Anthropic released Google Docs, Sheets and Slides connectors in beta at the same time, so Claude can also create and edit Google files from a Claude conversation. The add-on installs from the Google Workspace Marketplace and runs in Chrome, Edge and Safari.
Chandni Melwani
Founder & Editor
Oct 7, 2026 · 2 MIN READ
Know More
- — Plans: Pro, Max, Team and Enterprise, per Anthropic's Help Center. The user also needs a Google account that is allowed to install Google Workspace Marketplace apps.
- — Reach: the add-on can access only the file it is opened in, plus any connectors the user turns on. The separate Docs, Sheets and Slides connectors, used from the Claude app, can reach any file the user's Google Drive permissions allow.
- — Always asks: inserting an image from a web address, adding a link or removing anyone's access needs approval in either mode, with no "always allow" option. Claude cannot share a file, grant access or change its owner.
- — Beta limits, among others: no Firefox support; a six-minute cap on any single operation; charts it adds in Slides are inserted as images that do not update; Connected Sheets tabs backed by BigQuery cannot be read; chat history is stored per file in the browser and does not sync.
- — Data: Anthropic says content is not used for training and is deleted within 30 days, with exceptions. The extension does not inherit an organisation's custom retention settings, has no zero-data-retention option and is not covered for HIPAA.
Every change Claude makes from its new sidebar in a Google Doc is recorded in the file’s version history under the name of whoever asked for it, according to Anthropic’s Help Center. Anthropic opened that add-on, which also works in Sheets and Slides, to all paid Claude plans as a public beta on October 6. Collaborators see Claude’s edits “attributed to you, but not your chat.”
A copywriter who rewrites a brief with the add-on is, in that file’s history, the person who rewrote it. In Docs, Claude cannot read, reply to or resolve comments, so feedback a reviewer leaves in the margin has to be pasted into the sidebar first. Nor can it accept or reject existing suggestions, and its own changes go in as direct edits rather than suggestions. It can also propose changes as cards in the sidebar, which the user applies or dismisses one at a time.
Claude for Microsoft 365 goes further in two places, according to the same Help Center. Handing a task between apps is a Microsoft 365 feature, and the Microsoft version supports Amazon Bedrock, Google Cloud Vertex AI and Azure AI Foundry, where the Google extension, for now, requires a Claude account sign-in. The Google add-on arrives in a suite whose owner already sells companies its own assistant by the seat.
On Enterprise plans, Anthropic’s Compliance API makes the sessions from the add-on available, down to the edits Claude made, which gives an organisation’s admins a record of which edits were Claude’s. Inside the file, each one is, in Anthropic’s words, “a normal edit in the file’s history.” Slack put AI coding agents in channels colleagues can read. In Google Docs, Claude’s work reads as yours.
Related
Room for Disagreement
Anthropic says several of the beta's limits are deliberate, because the permissions that would lift them would also widen what the extension can access. By Anthropic's count the add-on asks Google for two permissions, one to show a sidebar and one to read and edit the file it is opened in. Google's Marketplace listing also shows it asking to see the user's primary email address and profile information. The add-on does not request access to Drive, Gmail, Calendar or contacts, or the ability to make network requests from the file. For a team that handles other companies' documents, an assistant that cannot see the next file over is the more defensible default, and a person can close the comment gap by pasting.
Frequently Asked Questions
Does a Google Workspace admin have to approve it?
It depends on the organisation's Google Workspace Marketplace settings. Anthropic's guide says many organisations block Marketplace installs by default. In that case a Workspace admin can allowlist Claude so people install it themselves, or a Workspace super administrator can install it for everyone, for chosen groups or for organizational units. On Team and Enterprise plans, an owner or primary owner of the Claude organisation also has to switch on the Docs, Sheets and Slides connectors before anyone can edit Google files from inside Claude.
Can a document instruct Claude to leak it?
Anthropic says files can carry instructions written by someone else, sometimes hidden in white text or tiny fonts. It says Claude treats document content as data, labels hidden text it finds, and is designed not to follow a document's instruction to share the file, fetch a link or contact anyone. Anthropic still tells users to review Claude's proposed actions, especially in files received from outside their organisation.
How is this different from Claude's existing Google connectors?
The existing Google Workspace connectors let Claude search Drive, Gmail and Calendar from claude.ai. The add-on works the other way round: Claude opens inside one Docs, Sheets or Slides file and edits that file. The new Docs, Sheets and Slides connectors are a third piece, letting Claude create and edit Google files from a Claude conversation.
Sources
- Anthropic, "Claude now works with Google Docs, Sheets, and Slides" (October 6, 2026): the public beta on all paid plans, the edit modes and the new connectors
- Claude Help Center, "Use Claude in Google Docs, Sheets, and Slides": plans, attribution in version history, limits, the permissions Anthropic says the add-on requests, admin install, data handling and Enterprise controls
- Claude for Google Workspace product page: browsers supported and what the add-on can access
- Google Workspace Marketplace listing for Claude: the permissions Google shows at install, including the user's email address and profile information
Chandni Melwani
Chandni Melwani is the founder and editor of New in AI, covering AI agents, M&A, and enterprise adoption. She holds a Master's in Management of Artificial Intelligence from Queen's University and brings a practitioner's perspective from her work in Data and AI leadership.
Follow ↗AI at Work Briefing
Get the five-beat briefing in your inbox.
The big story, frontier labs, AI at work, e-commerce and the capital crossing borders, each with the part that tells you why it matters, not just that it happened.
Free · No spam · Unsubscribe in one click
More from AI at Work
Jobs
Anthropic reports juniors fell to 9% of its technical hires
Anthropic's October 16, 2026 report on its own workforce says junior employees' share of its technical departments slipped from 21% to 14% since early 2025. It says the proportion of junior technical hires fell from 27% to 9% as hours worked by AI agents rapidly expanded relative to employees'.
Chandni Melwani · US · Oct 21, 2026
Jobs
Five Fed districts report some firms hiring less after AI
Some firms in five Fed districts reduced staff through attrition, delayed hiring or expected less job creation after AI investment raised productivity, the October 2026 Beige Book says. Most districts described little or no net change in overall employment levels.
Chandni Melwani · US · Oct 20, 2026
Workflows
Sophos says AI closes 52% of its MDR cases
In an October 9, 2026 OpenAI customer story, Sophos's technology chief says agents developed using OpenAI's Daybreak models automate about half the cases Sophos handles. In May, Sophos said its analysts had shifted to threat hunting, investigation, customer advisory and governance of the autonomous systems themselves.
Chandni Melwani · Oct 18, 2026