Deepfake fraud: what happens when a video call can be faked

In January 2024 an employee in the Hong Kong office of the engineering firm Arup received a message that appeared to be from the company's UK-based chief financial officer asking for a confidential transaction. They joined a video call that appeared to include the CFO and other staff, but the other participants were AI-generated. Police said the deepfakes were built from publicly available footage. The employee made 15 transfers totalling HK$200 million (about US$25.6 million) to five Hong Kong bank accounts. Arup reported it to police that month, police made the case public in February, and Arup confirmed it was the victim in May 2024. It said its internal systems were not compromised and its financial stability was not affected.

Nothing was hacked in the usual sense. The attackers did not break into a network; they impersonated authority, which is cheaper and works on people instead of software.

• It borrows authority. A request that appears to come from the CFO, on a video call with colleagues, in a confidential matter, is built to override the employee's own doubt.
• The raw material is public. Interviews, conference talks and social video give enough footage and audio to clone a likeness. Executives are the easiest targets because they are the most recorded.
• Controls assume identity is proven by a face. Many approval processes were written when seeing or hearing someone was proof.
• Payment rails are fast and often irreversible. Once funds leave to a mule account they are hard to claw back, and faster still if sent by instant or crypto rails, a point we return to in the ransomware economy.

The same capability scales in other directions: fake customer-support calls in a relative's cloned voice, fake job candidates, and fabricated clips during elections or a crisis. The harm to civilisation is less any single scam than the slow erosion of default trust in audio and video, which is a public good that took generations to build. When anything can be faked, everything can be denied.

1. 01Treat urgency, secrecy and unusual payment instructions as the warning sign, whoever appears to be asking.
2. 02Call back on a number you already hold, not one supplied in the message or on the call.
3. 03Require two people to approve payments above a threshold, and for any change of beneficiary details.
4. 04Agree a code word with family and key colleagues for emergency requests.
5. 05Do not rely on “detecting” a deepfake by eye. Artefacts like odd blinking are disappearing; process beats perception.

The same logic applies to machine-to-machine payments. A payment instruction should be proven by something an impersonator cannot copy: a cryptographic signature, not a face. That is why signed requests and verified webhook signatures matter more as faking gets easier.

• British engineering giant Arup revealed as $25 million deepfake scam victim — CNN (16 May 2024)
• Arup deepfake fraud scam victim, Hong Kong — Fortune (17 May 2024)