Claude AI Gains Autonomous Email Sending in Gmail Without User Approval
Anthropic enabled Claude to send, reply to and forward emails in Gmail without requiring user approval for each action, though approval remains the default setting and can be re-enabled. The feature raises concerns about an AI system taking autonomous actions on behalf of users without human review at the point of action.
What happened
Anthropic enabled Claude to send, reply to and forward emails directly in Gmail without requiring systematic user validation before each action. Previously, Claude could only draft responses that users had to manually copy-paste into Gmail.
User approval remains enabled by default, but can be disabled by individual users on paid plans. On Team and Enterprise plans, organization administrators control whether members can disable the approval requirement.
Reported impact
The shift from an assistant that suggests actions to one that executes them autonomously raises concerns about risks when an AI system acts without human review at the point of action. Time savings for repetitive tasks like follow-ups and routine customer responses come with the potential for unreviewed messages to be sent on a user's behalf.
Governance controls
Approval is the default setting and remains available for users who want to review messages before sending. On Team and Enterprise accounts, administrators can set policy on whether approval can be disabled.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.