
A finance employee at the engineering firm Arup attended a video call with the company's chief financial officer and several colleagues. He watched their faces move and heard their voices, and he authorised a transfer of 25.6 million dollars. Every other participant on that call was generated. The fraud surfaced only after the money was gone.
The employee did not fail a training module. He applied the verification standard almost every organisation still teaches, which is to confirm an unusual request with a person you can see and hear. That standard is now obsolete, and most payment approval processes have not been rewritten to say so.
What survives is verification through a separate channel the attacker does not control: a callback to a number held on file rather than one supplied in the request, dual authorisation that cannot be satisfied by one person on one call, and thresholds that force a delay no live pressure can compress.
This is the clearest example of why AI risk is not only an IT matter. The control that failed sits in finance, the loss sits on the balance sheet, and the fix is a change to an approval procedure. Governance that stops at the technology function will not reach it.
Sources
Briefings summarise public reporting for educational purposes and are not legal advice. Editorial policy