The $25 Million Video Call That Fooled Everyone
In January 2024, a finance worker at Arup - the British engineering firm behind the Sydney Opera House, the Beijing Bird’s Nest, and countless other landmark structures - received an email from the company’s chief financial officer.
The CFO was based in the UK. The employee was in the Hong Kong office. The email described a confidential transaction that needed to be processed urgently.
The employee was suspicious. It looked like a textbook phishing attempt. He’d seen the training. He knew the warning signs.
So he did what any security-conscious professional would do: he agreed to a video call to verify the request.
That decision cost Arup $25 million.
Everyone on the call was fake
On the video conference, the employee saw and heard the CFO. He also saw several colleagues he recognised - other senior members of the firm. They discussed the transaction. They answered his questions. Their faces looked right. Their voices sounded right. The meeting felt entirely normal.
The employee’s initial scepticism dissolved. He processed 15 separate transfers, totalling $25 million, to five bank accounts in Hong Kong.
He only discovered the fraud when he followed up with Arup’s actual headquarters. No one there had any knowledge of the meeting or the transactions. Every participant on that video call - except the victim - had been an AI-generated deepfake.
How it was done
Hong Kong police determined that the scammers built their deepfakes using publicly available video and audio of Arup’s executives - footage from online conferences, company meetings, and public appearances. The AI models analysed facial movements, speech patterns, and vocal characteristics to produce convincing real-time imitations.
This wasn’t a crude face-swap. It was a coordinated, multi-person, real-time deepfake deployment - arguably the most sophisticated social engineering attack ever documented at the time.
Arup’s CIO Rob Greig later reflected on the incident with unusual candour. Speaking to the World Economic Forum, he said: “What happened at Arup - I would call it tech-enabled social engineering.” The attackers didn’t compromise any systems or steal any data. They simply pretended to be people they weren’t, and the technology made the pretence convincing enough to work.
Curious about how accessible the technology had become, Greig tried to make a deepfake video of himself. Using open-source software, he managed it in about 45 minutes. It wasn’t perfect - but it was enough to demonstrate how low the barrier had dropped.
Why “get on a video call” is no longer enough
For years, the standard escalation path for a suspicious email was: verify on a video call. If you can see the person, you can trust the request. That assumption is now broken.
The Arup case is instructive precisely because the employee followed best practices. He was sceptical of the email. He sought visual verification. He participated in what appeared to be a legitimate multi-person meeting. The deepfake technology exploited his good instincts - the more cautious you are, the more powerful the confirmation becomes when the fake clears your checks.
This is what makes deepfake fraud fundamentally different from traditional phishing. It doesn’t exploit carelessness. It exploits diligence.
A pattern, not an anomaly
The Arup case was the most financially damaging, but it wasn’t isolated. The CEO of WPP - one of the world’s largest advertising companies - was targeted with a cloned voice on a fake Teams-style call. A UK energy firm lost €220,000 after an employee received a call from someone who sounded exactly like the company’s CEO directing them to pay a “trusted supplier.”
Across the financial sector, deepfake incidents surged 194% in Asia-Pacific in 2024. More than 10% of banks surveyed have lost over $1 million each to a deepfake call. Global deepfake fraud losses exceeded $200 million in Q1 2025 alone.
The solution is older than the technology
Every security expert, every law enforcement agency, every post-incident analysis arrives at the same recommendation: establish an out-of-band verification mechanism. Something the other person knows, not something they look or sound like. A code word. A challenge phrase. A shared secret.
The FBI’s own PSA is direct: “Create a secret word or phrase with your family members to verify their identities.” The FTC has launched dedicated initiatives to combat voice cloning. Cybersecurity researchers consistently recommend code words as a primary defence.
It’s the right answer. But it’s also an answer that, in practice, very few organisations implement rigorously. Static code words get forgotten, shared too widely, or bypassed under pressure.
This is why we built TrustWord. It takes the universally recommended approach - a shared verification phrase - and makes it cryptographically robust, automatically rotating, and impossible to forget. Every pair of people in a circle has a unique passphrase that changes every 2.5 minutes. Generated on-device. No server involved. No internet needed.
In the Arup scenario, one question would have stopped the entire attack: “What’s your TrustWord for me?” The real CFO would open the app and answer. A deepfake can’t.
Sources: CNN, Fortune, Financial Times, World Economic Forum, Hong Kong Police, CFO Dive
