Everyone else on the video call was fake. He paid out $25 million.

A staff member in the Hong Kong office of Arup, the British engineering firm, received a message appearing to come from the company's UK-based chief financial officer about a confidential transaction [1][2].

The request was unusual enough to raise doubts, so it was followed by a video conference. On the call were the CFO and other colleagues the employee recognised. Every one of them was an AI-generated fake, built from material of Arup executives [1][2].

Reassured by the call, the employee made a series of transfers to five different Hong Kong bank accounts, totalling HK$200 million — roughly US$25 million [1].

Arup reported the fraud to Hong Kong police in early 2024. It was not publicly identified as the victim until May that year, when it confirmed to the Financial Times that false voices and images had been used, declining to give further details [1][2].

Note what did not happen: nothing was hacked. No system was breached, no password stolen, no malware involved. The entire attack ran on the assumption that seeing and hearing someone is the same as knowing who they are [1][2].

The generative video and voice models behind this are the same ones doing genuinely valuable work — dubbing a film into forty languages, giving someone who has lost their voice a synthetic one built from old recordings, letting a small team produce training material they could never have filmed. The technology has no opinion about which it is doing. What changed is not that a new kind of villain appeared, but that a cheap, general capability arrived in a world whose verification habits quietly assumed that capability did not exist [1][2].

For most of living memory, a face and a voice on a live call were sufficient proof of identity — not because anyone reasoned it through, but because faking them in real time was too hard to be worth it. That assumption was load-bearing in a great many business processes, and it stopped being true without anyone updating the processes. The employee here was not careless. He did the thing careful people do: he was suspicious of an email, so he escalated to a richer channel to check. That is exactly the move that used to work [1][2].