Weaponized Email AI Assistants Could Help Attackers Hijack Accounts

Researchers demonstrate how attackers could abuse built-in email chatbots to evade detection, impersonate trusted employees, compromise executive accounts, and facilitate financial fraud. The post Weaponized Email AI Assistants Could Help Attackers Hijack Accounts appeared first on SecurityWeek .

The article describes how attackers can abuse built-in email AI assistants (e.g., Copilot-like agents embedded in mailboxes) once an account is compromised, using the assistant as a Living off the Land mechanism to discover sensitive information, impersonate employees, and escalate to executive account takeover and fraud.[1][8][13] Research on Email Agent Hijacking and real-world attacks like EchoLeak show that crafted emails and inbox content can override an assistant’s system prompts, turning routine summarization or reply features into stealth phishing, data exfiltration, and business email compromise channels.[11][15][19] From a RealGround perspective, this is a textbook indirect prompt injection and AI agent abuse risk on email-integrated assistants: organizations need continuous AI red teaming to simulate these inbox-driven hijack scenarios, validate that assistants resist hidden instructions in email bodies/HTML, and ensure defensive controls around account compromise, MFA, and assistant permissions are tested under realistic attacker use of the AI’s own capabilities.[8][15] Practically, security teams should map which email accounts are AI-enabled, constrain assista