Claude has started sending emails directly.
A feature that stopped at reading and drafting now extends to actually sending.
From reading and drafting to sending
Anthropic announced it on its official X account at 5:06 a.m. KST on August 19, 2026.
According to the official post, Claude can now write replies in Gmail and send them as is, and manage Google Drive files directly.
Previously, it supported only reading emails and drafting replies.
Anthropic's support documentation said until recently that "sending is not enabled."
This update removes that boundary.
Auto-send stays blocked until admins turn it off
If you ask, "Reply to Maya and send it," Claude reads the thread, writes the reply, and sends it right away.
Forwarding is also supported.
Approval before sending is on by default.
However, team and enterprise admins can turn off this approval step to allow automatic sending.
Whether to automate without repeated approvals is a decision to make as an organization-level policy.
The feature is available on all paid plans.
Connect Gmail or Google Drive from the connectors menu to get started.
Three things admins should check to avoid misdirected emails
Once you turn approval off, your organization bears the full risk of a misdirected email.
A wrong recipient, an inaccurate summary, or a reply that misses context can go out as is.
There are three things to check before rolling it out.
First, agree at the team level on how far to automate the approval policy.
Next, review the Gmail and Drive permission scope in the connectors menu.
Finally, it's safer to keep approval on for the first few days, check the quality of the emails Claude actually writes, and then widen the scope of automation.
Not officially confirmed: Anthropic hasn't published specific safeguards against misdirected emails, such as recipient restrictions.