Cybercriminals can transform Microsoft Copilot into a powerful tool for accelerating corporate account compromises and financial frauds. A controlled simulation showed how the assistant helped compromise a director's account and redirect a bank transfer worth US$ 247,500.
The attack starts after an intruder obtains access to an employee's email account. Using a simple command in Copilot, the attacker can create inbox rules that automatically send suspicious login alerts to the trash folder, significantly lowering the chance of detection.
The assistant then analyzes messages, documents, and internal relationships to identify executives and relevant conversations. This capability allows attackers to locate privileged targets quickly without manually reviewing thousands of emails.
Copilot can also mimic the victim's writing style and generate a phishing message based on real dialogue. In the test, the email directed the executive to a page that intercepted the session token and bypassed multi-factor authentication.
With the director's account under control, researchers asked the assistant for a summary of pending invoices and payments. Within seconds, the AI identified a US$ 247,500 transfer awaiting approval. A new message, written in the executive's usual tone, instructed the finance team to change the bank account details.
The case does not involve any specific vulnerability in Copilot but rather the abuse of permissions granted to an already compromised account.
