The Line Anthropic Just Crossed: From Reading Your Inbox to Acting On It
Until August 2026, Claude's Gmail relationship was read-only: back in April 2025, the assistant gained the ability to read Gmail [1], and even leading up to this update it could only check emails and summarize them - it could not respond [2]. That gap is now closed - Claude can draft and send emails, reply to threads, and forward messages directly, and on the Drive side it can search, share, move, and delete files without a human doing the clicking [4]. Anthropic's own framing is blunt about the goal: connect Gmail so Claude can 'search your emails, surface the information you need, and handle replies - without manually copying or forwarding anything.' [3]The guardrail is approval: by default Claude asks before it sends, replies to, or forwards anything, though Team and Enterprise admins can let members skip that prompt entirely [4]. That admin-level toggle is worth pausing on - it means the default safety behavior most users read about in headlines is explicitly designed to be turned off at scale, not a fixed rule Claude always follows.



