Researchers Compromised One Junior Employee's Inbox. The AI Assistant Did the Rest
New Research Email Fraud August 2026 · 8 min read In a controlled exercise, a security research team set out to get from an ordinary staff account to the chief executive's mailbox without being caught — using nothing but the AI assistant already built into the email system. The assistant hid the security alerts, mapped the organisation, learned the executive's writing style, found a pending wire transfer, and drafted the message that redirected it. Every step was a legitimate feature working exactly as designed. Over the past two years, almost every business email system has quietly acquired an AI assistant. It summarises long threads, drafts replies, searches your history in plain English, and sets up rules for you. For most small businesses it appeared as part of a subscription they already paid for — switched on by default, useful immediately, and never once discussed in a security review. A security research ...