Claude can now send Gmail messages on its own – the confirmation step can be switched off

Anthropic has given Claude the ability not merely to draft email in Gmail but to actually send it. The assistant can reply to messages, forward them, and compose and dispatch new ones. That shifts the model's role from preparing work to acting independently inside the mailbox.

The decisive detail is the default. Out of the box, Claude asks before anything goes out. That confirmation can be disabled, however – and anyone who does so allows the assistant to send in their name without a further prompt. The capability is available on paying plans.

In Team and Enterprise environments the choice does not rest with the individual user. There, administrators control whether staff may enable automatic sending at all. For organisations that is the more consequential setting, because a mistakenly sent message has an external effect and cannot be recalled.

Technically the step is less novel than it is organisationally significant. Assistants have been able to produce drafts for some time; what changes is who presses the final button. That is precisely where it is decided whether a model's error stays a correctable passage of text or becomes a message sitting in someone's inbox.

For users the practical lesson is that the convenient setting and the safe setting are not the same one here. Anyone switching on automatic sending should do so deliberately and for clearly bounded mailboxes – not as a blanket default across all correspondence.