Overview
- Anthropic updated Claude’s Google Workspace connector on Aug. 18–19 to allow the assistant to draft, reply to, forward and send Gmail messages when a user asks it to act.
- By default Claude asks for user approval before sending or forwarding an email but users can disable that prompt and Team or Enterprise owners can centrally allow actions to run without per-message approval.
- The send/reply/forward capability is limited to paid Claude plans and connector limits include no direct reading of Gmail attachment contents and access only when a user requests it.
- Anthropic says connector data from Gmail, Drive and Calendar will not be used to train its models, but reporters and users flagged risks from hallucinations and a reported Chrome-extension flaw that could let rogue extensions read data or trigger actions.
- The change promises time savings for heavy email users and teams but also raises operational risks for workplaces, so organizations are advised to test settings, use admin controls and keep approval on until reliability is proven.