Engadget reports that Claude can now manage Gmail inboxes, sending, replying to, and forwarding emails without user approval if permitted. The article outlines risks including hallucinated information, misinterpreted requests, and prompt injection attacks, which it says have actually occurred; OpenClaw previously deleted emails belonging to Summer Yue, an AI security researcher at Meta Superintelligence Lab. Mitigations include keeping the "ask before sending" approval setting active and giving specific instructions, though prompt injection cannot be fully prevented.
No score is assigned. Sources and their independence are shown in the citation chain below.