The finance worker almost didn’t join the call. That email had felt off: a message from the company’s UK chief financial officer about a “secret transaction,” the kind of note that sets off every fraud alarm a trained employee has. So the scammers did the one thing that could dissolve the doubt. They invited him to a video conference with the CFO and several colleagues he recognized.
He joined. He saw their faces. He heard their voices. Reassured, he made 15 transfers over the following days, about $25 million in total, to five bank accounts. Every single person on that call, except him, was a fake.
Key Takeaways
- A finance worker at Arup’s Hong Kong office wired roughly $25 million (HK$200 million) to fraudsters in January 2024 after a video call with what he believed were his colleagues.
- Every participant on the call except the victim was an AI deepfake, generated from real audio and video of Arup staff, including a convincing likeness of the UK-based CFO.
- The money left in 15 separate transfers to five different Hong Kong bank accounts before the fraud was uncovered.
- The victim was suspicious at first. An initial email about a “secret transaction” raised a red flag; the live video call is what overrode his instinct.
- As of late 2025, no arrests, no suspects named, and no money recovered. Arup, a firm behind landmarks like the Sydney Opera House, confirmed itself as the victim in May 2024.
How Did the Arup Deepfake Scam Work?
It worked by turning the employee’s own caution into the trap. The scam began with a phishing email in the name of Arup’s UK chief financial officer, describing a confidential transaction that needed to be handled quietly. He later said he suspected it was a phishing message. Under normal circumstances, that suspicion is the whole defense.
So the attackers escalated to something that felt unfakeable: a group video call. Using publicly available footage and audio of real Arup executives and staff, they generated live deepfake avatars, faces and voices that matched people the employee knew. On the call, the “CFO” and several “colleagues” walked him through the urgency and the instructions, and seeing familiar faces and hearing familiar voices did exactly what the attackers needed: it converted a suspicious email into a legitimate-feeling order from the boss.
Over the next several days the worker made 15 transfers totaling around $25 million to five Hong Kong accounts. It surfaced only when he later checked with Arup’s head office. By then the money was gone. This is social engineering at its most advanced, but the underlying move is ancient: manufacture authority and urgency, then remove the moment where someone might stop to verify. It is the same human exploit that let a single phone call shut down MGM’s casinos, just dressed in far better technology.
Why Didn’t Anyone Catch It?
Because every checkpoint that should have caught it was designed for a world where you can trust your own eyes and ears. Corporate anti-fraud training tells employees to be wary of email requests and to confirm unusual payments with the person who asked. He did grow wary of the email. What tripped him was the thing he confirmed it against: a video call, which felt like the highest possible proof of identity and was in fact the forgery itself.
Nothing in the process required an independent channel. There was no mandatory callback to a known number, no second approver who had to sign off separately, no code word agreed in advance. Verification and attack were the same event. When the proof of authenticity is a face on a screen, and faces can be synthesized in real time, the whole control collapses, a more sophisticated cousin of the failures across our hacks and breaches archive, where the technology holds and the human trust layer is what gets picked.
| Red flag present | Why it failed to stop the transfer |
|---|---|
| Suspicious email about a “secret transaction” | Employee was reassured by the follow-up video call |
| Unusual, urgent, high-value request | Urgency and secrecy were framed by a trusted “CFO” |
| Payment to five unfamiliar accounts | No independent callback or second-approver rule required |
| 15 separate transfers over days | No out-of-band check was triggered by the pattern |
Is Deepfake Fraud a Growing Threat?
Yes, and Arup is the landmark case precisely because of its scale and who the victim was. This was not a careless startup. Arup is a global engineering firm behind some of the most famous structures on Earth, with sophisticated controls. If its finance function could be fooled, most can. Arup’s own regional leadership said afterward that the number and sophistication of these attacks, deepfakes, voice cloning, and synthetic identities, was rising sharply.
Tooling has only gotten cheaper since. What once required a studio now runs on consumer hardware, which means the barrier is no longer technical skill but a willingness to try. That same collapse of “seeing is believing” runs through modern fraud, from synthetic-identity crime to the way criminals now openly recruit insiders and bribe support staff to reach customer data. Arup simply showed the most cinematic version: an entire meeting that never happened.
The Critical Choice
The decision that made the loss inevitable was treating the video call as proof of identity. Every other element of the attack was survivable: a suspicious email got correctly flagged, the urgency was a known red flag, and the unfamiliar accounts were right there to see. What defeated all of it was the unspoken assumption, baked into the company’s payment process and into human instinct itself, that if you can see and hear your CFO on a live call, it is really your CFO. In an era of real-time deepfakes, that assumption is a security hole. The fatal choice was not a single click by one overwhelmed employee; it was the organizational decision to let visual and audio presence stand in for verification, with no independent channel required for a $25 million transfer. The worker who trusted his eyes was doing exactly what he had been trained to do. That training was written for a world that no longer exists.
Where Things Stand Now
More than a year after the transfers, the case remains unsolved. Hong Kong police have made no announced arrests in the Arup matter, no perpetrators have been publicly named, and the roughly $25 million has not been recovered. Arup has used its own experience to warn other companies, and the incident has become the reference example in boardrooms and security briefings worldwide for why “verify the request” now has to mean verify it on a channel the attacker cannot fake. Deepfakes are only getting better. The defense, oddly, is getting older: a phone call to a number you already trust, a second human who has to agree, a question only the real person could answer. This page will be updated if arrests are made or funds are recovered.
