Hong Kong deepfake scam: how a deepfake CFO on a video call cost Arup $25 million

Quick answer In early 2024, a finance employee at Arup in Hong Kong joined a video call where the "CFO" and other colleagues were all deepfakes. Believing the call was real, the employee made 15 transfers to five local bank accounts totaling HK$200 million, about US$25.6 million. Arup confirmed in May 2024 that fake voices and images were used.

What happened on the call?

It starts with a message. It claims to come from the company's UK-based chief financial officer, and it talks about the need for a secret transaction.¹ The finance worker in Hong Kong is suspicious. It looks like a phishing email.

Then comes the invitation: a video conference with several participants.² The worker joins. On screen are faces they know. The voices sound right. The CFO is there, and so are other members of staff. CNN reported, citing Hong Kong police, that the worker put aside the early doubts because the others on the call had "looked and sounded just like colleagues he recognized."¹

None of them were real.

"(In the) multi-person video conference, it turns out that everyone [he saw] was fake," senior superintendent Baron Chan Shun-ching told the public broadcaster RTHK.¹

Following instructions from the call, the worker made 15 transactions to five local bank accounts. The total came to HK$200 million.² CNN put that at about US$25.6 million.¹ Only afterwards did the employee check with the company's head office, and that is when the scam came to light.¹²

Police said they received the report on 29 January 2024 and classified the case as "obtaining property by deception," handled by the cybercrime unit.² At first the company was not named. In May 2024, London-based Arup, the firm behind the structural engineering of the Sydney Opera House, confirmed to CNN that it was the victim.³

Unfortunately, we can't go into details at this stage as the incident is still the subject of an ongoing investigation. However, we can confirm that fake voices and images were used.

Arup spokesperson, statement to CNN, May 2024

Why did the scam work?

The worker did the right thing at first. The message looked like phishing, and the worker treated it that way. What broke the doubt was not the email. It was seeing and hearing familiar people on a live call.

Police believe the fraudsters prepared in advance. "I believe the fraudster downloaded videos in advance and then used artificial intelligence to add fake voices to use in the video conference," Chan said, as quoted by the Guardian.² CNN reported that fraudsters in several recent cases were believed to have used deepfake technology to modify publicly available video and other footage.¹

The setup was built for pressure: a request framed as confidential, coming from someone senior, confirmed by a group of colleagues all at once. A single fake face might raise questions. A room full of them felt like proof.

A few caveats matter. Arup has not shared the full details because the investigation was ongoing, and police did not say exactly which tools were used. Arup also said its "financial stability and business operations were not affected and none of our internal systems were compromised."³ In other words, nobody hacked Arup's systems. The fraud targeted a person and a payment process.

Arup's global chief information officer, Rob Greig, said the company faces regular attacks, "including invoice fraud, phishing scams, WhatsApp voice spoofing, and deepfakes," and that "the number and sophistication of these attacks has been rising sharply in recent months."³ If a firm with 18,500 employees across 34 offices can be hit this way,³ a small brand should assume it can be too.

What this means for your business

You do not need to be a multinational to face this. Small teams often move money on a founder's word, sent over chat or a quick call. That is exactly the pattern this scam copied. A few habits help:

• Treat "secret" and "urgent" as red flags, not reasons to skip steps. The Arup request was framed as confidential. Real urgency still survives a two-minute check.
• Verify through a second channel you already trust. Call the person back on a number you have on file, or message them on a channel the request did not come from. The Arup fraud surfaced only when the employee contacted head office, after the money had gone.
• Do not let a video call count as approval. Police said people "must be vigilant even in meetings with lots of participants."² Faces and voices can be faked; a payment rule cannot be talked out of.
• Split large payments across two people. One person asks, a different person approves, and neither can be the only check.
• Watch for unusual payment patterns. A run of transfers to unfamiliar accounts is the kind of thing a second set of eyes should catch before the money leaves.

The same lesson shows up in AI voice clone scams, where a familiar voice on the phone does the convincing. If you are bringing AI agents into your own finance or operations work, the question is the same: who has to say yes before money moves? We wrote about that in why the confirm button matters.

How ZNOS handles this

The way we run it, payments are not something an AI employee or a single message gets to decide alone. In ZNOS, each AI employee owns a job (Raghu runs operations, Sophia is the executive assistant, Jason handles sales) and works from the company's live data through the Brain.

Autonomy is earned gradually. Routine decisions are handled, but risky ones, including anything involving money, customers, things published under your name or production changes, wait for a human approval. An Audit Log records who changed what, so there is a trail if something looks off later.

The Brain also watches orders, invoices and shipments around the clock and raises plain-English alerts in the app, by email or on WhatsApp when something looks wrong. It does not replace a call-back rule for big transfers, and we would not want it to. It is one more set of eyes on the money. You can meet the ZNOS team to see who owns which job.

For more cases like this one, see our hub of real AI incidents.