Claude KI now sends emails independently in Gmail

Anthropic adds autonomous capabilities to Claude AI in Google Workspace. The model now composes and sends emails in Gmail completely independently. Manual confirmation of the actions is no longer absolutely necessary.

Autonomous AI in the email inbox

The AI ​​company Anthropic has announced that it is significantly expanding the functions of its assistant Claude for Google Workspace. From now on, the language model can not only write drafts, but also actively send, reply and forward messages in Gmail. At the same time, the software receives extended access rights for Google Drive in order to directly manage documents in the users’ everyday work.

Until now, the integration was limited to searching mailboxes and formulating text suggestions. The current update transforms the system from a pure chatbot to an agent that can carry out advanced actions. But there are of course limitations: The innovations are only available to users of paid subscriptions and require prior activation of the corresponding connectors in the settings.

Users remain in control

According to information from 9to5Google Manual confirmation before executing an action is enabled by default. However, users also have the option of deactivating the protection function. In this case, the model processes and sends the mail completely independently, without the human reading the text first. For company accounts, administrators decide whether to release the automation.

In addition to email communication, the update includes extensive powers for cloud storage. The system can now search, share, move, or trash files in Google Drive. In the factory setting, the software also asks for permission before each write access. Authentication takes place via the user’s existing account, with the applicable access rights being exactly mirrored.

Functionality and data protection

The new functions may be practical, but they also involve risks: sending directly is less forgiving of errors than a mere draft. A misunderstood context in an incoming message will immediately result in a genuine response in automatic mode. However, people with high communication volumes save a lot of time when dealing with routine requests.

But you should always keep in mind that a draft that has not been sent can be corrected at any time, but a message that has already been sent can hardly be corrected. Anthropic emphasizes that the data retrieved from Workspace is not used to train the models. The company’s range of functions is primarily aimed at professional users who want to bundle many processes in everyday office life.