Anthropic has announced that its Claude assistant can now send emails in Gmail and manage files in Google Drive. Until now, it could search the inbox, summarize a long thread, or prepare a draft, but the final step—actually sending it—was left to the user. Now, all you need to do is ask it to reply to a specific conversation, and Claude will compose and send the response. Those who want to review the message first can keep automatic sending disabled.
It can read emails, move files, and schedule meetings
Wondering what else Claude can now do? In Gmail, it can search for and read messages, send, reply to, or forward emails. The assistant can also write and list drafts, work with labels, and access message metadata. This includes information about attachments, but not access to their actual contents.
Google Drive offers a similarly broad range of options. Claude can find, open, share, move, or delete documents. It can read spreadsheets, presentations, PDF files, images, and documents from the Office suite. It allows users to upload any file type with the option to convert it to Google formats, creates folders, displays access permissions and recent changes in Drive, and can save its own outputs directly to your storage. However, for all documents, the assistant processes text only. The system ignores images embedded in documents, as well as comments and suggestions.
In Calendar, the assistant displays events, including those in shared calendars, and can create, edit, or delete appointments. It can also find a mutually available time for invitees, manage the attendee list, respond to invitations, or create recurring meetings.
By default, Claude asks for confirmation before sending, replying to, or forwarding any message. With Team and Enterprise plans, however, the account owner decides whether members can perform these actions automatically. The same applies to sharing, moving, and deleting files in Drive. Anthropic also notes that permission to send email will now appear in the Google sign-in window.
What to watch out for
Tom's Guide highlights a risk that will likely be obvious to anyone who has ever sent an email by mistake. If you allow the assistant to send messages without review, the consequences of a potential hallucination could be highly unpleasant. It also points to a recent discovery by researchers who found a vulnerability in the Claude extension for Chrome through which a malicious extension could read data and perform actions in user accounts. Such an extension could then prompt the assistant to send a message without the mailbox owner knowing. The publication therefore recommends keeping approval enabled and testing in advance whether the resulting text sounds natural.
User reactions vary widely. TechRadar collected reactions from X, including the objection that actually sending an email was never the hardest part. The difficult part is the apology that may have to follow. Another user said they would rather rely on their own judgment when communicating with clients than on an assistant that could invent false information about prices or another customer's ownership.
Anthropic provides its own assurances. Claude can only see data from the account you connect to it, accesses that data only when you ask it to do something, and retrieves the minimum information necessary. It mirrors your existing permissions, so it cannot access anything you could not access yourself. The retrieved data is stored on the company's servers, remains linked to the specific conversation, and is deleted when the chat is deleted. According to the company, its models are not trained on content from Gmail, Drive, or Calendar.
The help documentation itself lists the feature's limitations. Claude cannot access the contents of attachments, does not support some of Gmail's more advanced filters, more complex queries require several API calls, Google's quota limits apply, and performance may decline with large mailboxes. So anyone expecting the assistant to go through a ten-year-old thread with invoices attached is out of luck.
Alongside this announcement, Anthropic expanded Claude Cowork to mobile and the web and extended the increased weekly limits for Claude Code through the end of August.
