Particle.news

Anthropic's Claude Can Now Send Emails From Gmail

The connector update hands Claude executable control of users' Gmail and raises fresh privacy and security questions.

Overview

  • Anthropic announced on August 18–19 that updated Google Workspace connectors let Claude draft, reply to, forward and send Gmail messages when a user connects their account.
  • By default Claude asks for human approval before sending, but Team and Enterprise plan owners can enable automatic sends so the assistant will dispatch messages without per‑message review.
  • The new features are gated to paid Claude plans and require users to enable the Gmail connector before Claude can access inbox content or act on a connected account.
  • Anthropic says connector data is accessed only when a user requests a task and that Gmail, Drive and Calendar connector data will not be used to train models, and the assistant cannot directly read attachment contents.
  • Security researchers' prior report of a Claude‑for‑Chrome flaw that let rogue extensions read data and take actions has renewed warnings that browser extensions or integrations could magnify risks if they can trigger sends without a user's knowledge.