Claude can now send Gmail emails on your behalf — without asking permission first

Full text

Claude can now draft and send messages on your behalf in Gmail

However it can do so without asking for permission, potentially leading to awkward responses

Claude also gets stronger Google Drive access, but says it won't train its AI on your files

Anthropic has revealed that Claude's integration with Gmail has taken a major step forward - with the AI now able to draft and send emails on a user's behalf.

"Ask Claude to reply to a thread, and it drafts and sends the response. You control when it needs your approval," the company wrote in an update on X.

However the news immediately prompted a flood of replies and comments from users worried about the potential implications of having an AI send emails unseen by them.

Ask for forgiveness, not permission?

Although a Claude connector has been available in Google Workspace for a while, actually sending an email on the user's behalf is a significant step forward, and a potentially troubling one at that.

Ultimately, the user will still have to ask Claude to perform the action first, but how it drafts the message, and the content within, has the potential to lead to some serious embarrassment for some users.

(Image credit: Anthropic) The tool, which is live now, will only be available to users on paid Claude plans, , and Anthropic later clarified the process in a support page , noting that asking for approval is the default setting, and on Team and Enterprise plans, admins will be able to decide "whether members can allow these actions to run without asking each time".

The update will also mean Claude can forward emails, and manage files in Google Drive, helping users track down the information they need with a single command, as well as uploading files, creating folders and add Google Docs directly to chats and projects.

"Claude mirrors your existing permissions—you cannot access information you don't already have access to in Google Workspace," the support page noted, adding that Anthropic would not, "train our models on your Gmail, Drive, or Calendar connector data, ensuring your private information remains private."

Users appeared divided on the new tool, with some supporting the launch, and others seemingly very opposed.

"sending the email was never the hard part, the hard part is the apology email it sends next," one X user wrote.

"I would rather use my own head than to let Claude reply especially responding to my clients," another said , "Don't want Claude hallucinating me into exposing my cost or another client's IP. Claude can apologize but not get your client back. Need to keep our brains working and not letting it totally rot."

sending the email was never the hard part, the hard part is the apology email it sends next August 18, 2026