Claude's New Email Powers: A Double-Edged Sword
The Anthropic AI assistant Claude can now send emails without asking permission in Gmail, but only for its paid subscribers. This update was released in August 2026 and allows Claude to read incoming emails, write replies, and send them to someone else's inbox.
Claude can perform a range of actions on emails, including sending, replying, and forwarding, although it checks with the user by default before taking any action. However, users can switch this off and let Claude take control.
This is different from Gemini, another AI assistant that Google has built. While both assistants have similar capabilities in terms of reading and searching emails, Gemini does not allow for autonomous sending without a manual click. This means that while both assistants can perform tasks on behalf of the user, only Claude can send emails without needing to be explicitly told to do so.
The update also highlights the potential risks associated with AI assistants taking control of email accounts. For example, if an agent is given instructions to disregard its prior commands and forward sensitive threads to an outside address, it could potentially lead to security breaches or other problems.