# Deepfake CEO Fraud: Lessons From Arup and Ferrari

> Deepfake CEO fraud uses a cloned voice or a fake video of a senior executive to push an employee into an urgent, secret payment. The defence is procedural: every payment request that arrives by call, chat or video is confirmed through a second channel the company controls, and no executive is exempt.

- URL: https://fake-ai.com/voice-and-scams/deepfake-ceo-fraud/
- Language: en
- Updated: 2026-10-19
- Publisher: Fake AI (OnlineBuilders LLC, https://onlinebuilders.biz)

January 2024: a finance employee in the Hong Kong office of the British engineering firm Arup joined a video call. The company's chief financial officer was on screen, along with several colleagues. They discussed a confidential transaction. During the following days the employee made 15 transfers to five bank accounts, about HK$200 million in total, roughly US$25 million. Everyone else on the call had been generated. Arup confirmed in May 2024 that "fake voices and images were used".

Six months later, a manager at Ferrari got a WhatsApp message from someone claiming to be CEO Benedetto Vigna, followed by a call in a convincing copy of Vigna's voice, southern Italian accent included. The manager asked which book Vigna had recently recommended to him. The caller hung up.

Same technique, opposite outcome. The difference was not better detection software. It was a habit.

## What deepfake CEO fraud is

CEO fraud is older than generative AI. Its classic form is an email that appears to come from the boss and asks an employee to make an urgent, confidential payment. Security teams answered with a simple rule: if in doubt, call the boss. Cloned voices and live video fakes attack exactly that rule, because now the call itself can be fake.

The FBI described the shift in December 2024. Criminals use generative AI to create "real-time video chats" in which they pose as company executives or authority figures ([FBI IC3](https://www.ic3.gov/PSA/2024/PSA241203)). Germany's BSI notes that face-swap models can run close to real time and need only a few minutes of good video of the target to train ([BSI](https://www.bsi.bund.de/EN/Themen/Unternehmen-und-Organisationen/Informationen-und-Empfehlungen/Kuenstliche-Intelligenz/Deepfakes/deepfakes_node.html)). Executives who give interviews, keynotes and earnings calls provide that material for free.

**The Arup case as reported**

- **Jan. 2024:** Employee receives a message from the apparent UK CFO about a secret transaction.
- **Jan. 2024:** Video call with the fake CFO and fake colleagues. 15 transfers to five accounts follow.
- **Later:** The employee checks with head office and the fraud comes to light.
- **Feb. 2024:** Hong Kong police describe the case publicly without naming the company.
- **May 2024:** Arup confirms it was the victim.

Source: Hong Kong police statements (Feb 2024) and Arup's confirmation (May 2024), as reported by the South China Morning Post

## Why the employee in Hong Kong believed it

According to early police statements, the employee was suspicious of the first message, which talked about a secret transaction. The video call removed that doubt. Seeing several familiar faces agree with each other is powerful social proof, and a junior employee is unlikely to challenge a CFO in front of colleagues.

The attack bundled the pressure points that fraud trainers warn about:

- authority: the request came from the top
- secrecy: the deal was confidential, so no one else could be asked
- urgency: transfers had to happen quickly
- consensus: other "colleagues" on the call went along with it

None of those depends on AI. The deepfake just made them believable over video.

## Why the Ferrari attempt failed

According to Bloomberg's account, the Ferrari manager noticed small things first. The WhatsApp messages came from a number that was not Vigna's usual one, with a different profile photo. The voice on the call had slight mechanical intonations. Then the manager did the decisive thing: asked a question only the real Vigna could answer, about a book he had recommended in person.

That is a verification step that costs nothing, does not require technical knowledge and works against any quality of deepfake. Ferrari declined to comment on the reports, so treat the details as reported rather than confirmed.

## Controls that work against a perfect fake

Assume the fake will be good enough. Build controls that do not depend on anyone noticing it.

**Payment controls against deepfake requests**

| | Stops a perfect deepfake | Cost to run |
|---|---|---|
| Call-back on a number from the internal directory | yes | low |
| Second approver for new payees and changed bank details | yes | low |
| Holding period for first payments to new accounts | yes | medium |
| Personal question only the real person can answer | partly | low |
| Staff training to spot artefacts | no | medium |
| Deepfake detection software on video calls | partly | high |
*Our assessment. "Stops" means the control works even if the fake voice or video is flawless.*

The core rule should fit on one line and apply to everyone, including the CEO: **no payment, change of bank details or release of confidential data on the strength of a call, chat or video alone.** Confirmation always goes through a second channel that the company controls, using contact details from the internal directory, never from the request.

Executives have to back this publicly. The fraud works because employees fear embarrassing a senior person. If the CFO says at an all-hands meeting "I will never be annoyed if you call me back to check", the social pressure disappears.

## A verification procedure you can adopt

**Five steps before any urgent payment request is executed**

1. **Pause** Urgency and secrecy are triggers to slow down, not speed up.
2. **End the contact** Leave the call or chat politely: I'll confirm and get back to you.
3. **Call back** Use the number in the company directory or HR system, not the one in the message.
4. **Second approver** A colleague who was not on the call signs off on the payment.
5. **Log it** Record the request and the check. Patterns across departments reveal campaigns.

Test the procedure. Some companies run internal exercises with a harmless fake request to see whether staff follow it. If you do, debrief without blame. The goal is a habit, not a list of people who failed.

## If the money has already left

Minutes count. Call your bank's fraud desk immediately and ask for a recall of every transfer, including ones that look completed. International payments sometimes sit in intermediary banks for hours. Then report to the police and, in the US, to the FBI's IC3, which runs a process for freezing funds in business email compromise cases.

Preserve everything: the meeting invite, the chat history, the phone numbers, the video platform's logs and the payment records. Platforms and banks delete logs on their own schedules.

Tell staff quickly and plainly what happened. Attackers who succeed once often try again with a variation, and a short internal alert is the cheapest way to stop round two. If the attack used a real executive's likeness, warn partners and customers too, because the same fake may be used against them next.

## Where technology helps and where it doesn't

Detection tools for video calls exist, and some can flag manipulated streams. They are worth evaluating for high-risk roles. But automated detectors generalise poorly to new fake-making methods, which the BSI points out explicitly, and a control that depends on software spotting the fake fails exactly when the attacker uses something new.

Better technology investments are boring: enforced approval workflows in your payment system, alerts when bank details change, and directory data that staff can reach quickly. Companies that want those checks built into their finance or chat tools can get help from a studio like [OnlineBuilders](https://onlinebuilders.biz/solutions/cyber-security).

For the human side, read our guide on [how to detect an AI voice](https://fake-ai.com/voice-and-scams/detect-ai-voice/) and the checklist on [how to spot a deepfake video](https://fake-ai.com/images-and-video/how-to-spot-deepfake-video/). The same call-back logic protects families too, as our article on the [AI voice cloning scam](https://fake-ai.com/voice-and-scams/ai-voice-cloning-scam/) shows. The first step for your company is short: write the one-line rule, have the CEO send it to everyone, and put it in the payment approval screen.

## FAQ

### How did the Arup deepfake scam work?

A finance employee in Hong Kong received a message that appeared to come from the UK-based CFO about a confidential transaction. On a video call with what looked like the CFO and other colleagues, all generated, the employee was instructed to make 15 transfers totalling about HK$200 million to five bank accounts.

### Is deepfake CEO fraud covered by cyber insurance?

It depends on the policy. Some policies treat payments an employee made voluntarily, even under deception, as social engineering losses with separate and often lower limits. Check the wording before you need it.

### Which departments are targeted most?

Anyone who can move money or change bank details: finance, accounts payable, treasury and assistants to executives. HR and payroll are targeted with requests to change salary accounts.

### Does a code word work in a company?

It can, but it scales poorly and leaks easily. A fixed call-back procedure using numbers from the company directory is more dependable, because it does not rely on anyone keeping a secret.

Photo: Project Kei, CC BY-SA 4.0 (https://commons.wikimedia.org/w/index.php?curid=90881588)