The CEO Deepfake on a Video Call That Cost $25.6M: The Protocol That Would Have Stopped It | inicIA
Direct answer
In January 2024, a finance employee at engineering firm Arup in Hong Kong transferred roughly $25.6 million (HK$200 million) across 15 wire transfers to 5 bank accounts, after a video call in which the company's chief financial officer and several colleagues —every one of them— were AI-generated deepfakes built from public material. Hong Kong police disclosed the case in February 2024; Arup identified itself as the victim in May 2024. The most unsettling detail: the employee had been suspicious of the initial phishing email, but the video call "with familiar faces and voices" turned suspicion into conviction. The lesson: seeing and hearing someone no longer proves they are real. The only antidote is an out-of-band verification protocol.
How it happened, step by step
- The employee received an email appearing to come from the UK-based CFO: a confidential, urgent, discreet transaction. The classic CEO-fraud (BEC) shape.
- The employee grew suspicious and asked to verify. He was invited to a video call with "the people involved."
- On the call he saw the CFO, recognized colleagues, the voices matched, and the conversation was coherent. It was all synthetic: the deepfakes had been built from public recordings of conferences and virtual meetings featuring the executives.
- After the call, he authorized the 15 transfers. The fraud was only discovered when he contacted the real headquarters and was told that meeting never happened.
As of early 2025, per reports, none of the funds had been recovered. And it was not the only attempt that year: Ferrari avoided a similar attack because an executive asked the supposed CEO a personal question the AI could not answer. The difference between the two outcomes was a moment of doubt and the willingness to act on it.
Why your company is exposed too
You do not need to be Arup. The problem's data: a University College London study found people correctly identify deepfake voices only 73% of the time, and dedicated training improves that by just 3.84%. The FBI has warned since 2022 about deepfakes in remote job interviews. FinCEN (US Treasury) issued a November 2024 alert to financial institutions about deepfake fraud. And a Surfshark study (April 2026) estimates documented global deepfake fraud losses of about $2.19 billion between 2019 and March 2026. Your executives have public voices and faces: that is the attack's raw material.
The anti-deepfake protocol (5 rules)
- Out-of-band verification, always. Every sensitive transfer or instruction arriving by voice or video is confirmed through a second, independent channel: a call to the executive's pre-registered number — never the number in the message.
- Verbal passphrase. Agree on a verification phrase with your team, asked on sensitive calls. It is the cheapest and most effective measure.
- Authorization limits by amount. No single person authorizes transfers above a threshold: mandatory dual sign-off, with no exceptions for "urgency."
- Urgency is the warning signal. Fraud lives on time pressure ("confidential," "right now"). Every instruction flagged as urgent gets reinforced verification, not reduced.
- Physical security keys. For critical access, hardware keys resist impersonation better than any visual check.
Related articles
- What to do when your AI agent makes a mistake
- Privacy and data: what happens to your customers' information
- Florida wants criminal liability for chatbots that help commit crimes
Frequently asked questions
Can this happen to a small business?
Yes, at proportional amounts: the attack does not depend on company size, but on having executives with public presence and employees who execute transfers. The protocol costs almost nothing.
Isn't deepfake-detection software enough?
Not as the sole defense: automated detection remains probabilistic and lags behind how fast generators improve. Procedure (out-of-band verification) is more reliable than detection technology.
Does cyber insurance cover this?
It depends on the policy: wire fraud is typically classified as social engineering, which many policies exclude or cap. Review your funds-transfer fraud sublimits today.
Should I ban video calls for authorizing payments?
Do not ban them: strip them of final authorization power. The video call informs; authorization is confirmed through the pre-registered channel.
Take the next step
If you handle payments, transfers, or sensitive instructions and want an anti-fraud protocol designed for the deepfake era, talk to inicIA on WhatsApp: https://wa.me/17866120930
By inicIA
