The Arup Deepfake Video Call: What Happened and What It Teaches

What Happened in the Arup Deepfake Scam?

An employee in the Hong Kong office of Arup, the British engineering and design firm, received a message that appeared to come from the company's UK-based chief financial officer about a confidential transaction. According to reporting of the Hong Kong police briefing, the employee suspected a phishing attempt. The doubt faded after a video conference in which the employee saw and heard what appeared to be the CFO and several other colleagues.

Following instructions given on and after the call, the employee made 15 transfers totalling HK$200 million, about US$25 million, to Hong Kong bank accounts. The fraud came to light when the employee later checked with the company's head office in the UK. Senior Superintendent Baron Chan Shun-ching of the Hong Kong police described it as a multi-person video conference in which, it turned out, everyone the employee saw was fake.

Timeline of the Arup Deepfake Case

The public record comes from two disclosures several months apart. Hong Kong police described the case without naming the company in early February 2024, and Arup confirmed its involvement to the press on 17 May 2024. Reports at that time said no arrests had been made and the investigation was continuing.

How Did the Deepfake Video Call Fool the Employee?

Three things made it work. The first was social proof. A single suspicious email is easy to question, but a call with several familiar senior people, all agreeing, is hard to doubt. The second was channel trust. Many companies use video calls to confirm who they are dealing with, and few finance procedures treat a video call as something that can be forged. The third was the setup. The opening message framed the transaction as confidential, which discouraged the employee from checking with anyone outside the call.

Arup's then chief information officer, Rob Greig, said when the company confirmed the case that the number and sophistication of such attacks had been rising sharply in recent months. Arup's statement made a point that matters for defenders: none of its internal systems were compromised. The criminals did not need network access. They needed enough public video and audio of executives to build convincing likenesses, and a payment process that a convincing call could open.

What the Arup Case Teaches About Deepfake CEO Fraud

Treat seeing and hearing someone as a weak signal for high-risk actions. A request to move money, change bank details or share credentials should trigger verification through a channel the requester did not choose: a call back to a number in the corporate directory, a message in an internal system, or an in-person check. The verification must be required by process, so that staff are not left deciding on their own whether a senior colleague looks real.

Specific controls follow from that principle:

• Require dual authorisation for transfers above a threshold, with the second approver verifying independently of the original request.
• Never accept confidentiality as a reason to skip verification; make that rule explicit for finance and treasury staff.
• Split large or urgent payments to new beneficiaries across a hold period long enough for a call-back.
• Agree out-of-band verification phrases or procedures for executives who approve payments remotely.
• Run deepfake detection on recorded or live media where the workflow allows, and treat a high-risk score as a reason to stop and verify.

How Regulators Responded to Executive Deepfake Fraud

The case became a reference point for financial regulators. FinCEN's alert FIN-2024-Alert004 of 13 November 2024 warned that criminals have reportedly used generative AI tools to impersonate executives or trusted employees and instruct victims to transfer large sums, citing press coverage of the Hong Kong video call among its references. It asked financial institutions to file suspicious activity reports with the key term FIN-2024-DEEPFAKEFRAUD and listed red flags including inconsistencies during live verification and media flagged by deepfake detection software.

The FBI's 2025 Internet Crime Report, published in April 2026, recorded business email compromise losses involving AI of more than $30 million and noted that voice cloning can be used to request wire payments. Those are complaint-based figures and undercount incidents like Arup's that are handled privately.

The response side matters as much as prevention. In the Arup case the fraud surfaced only when the employee checked with head office after the transfers. Companies should make it easy and blame-free for staff to raise doubts quickly, because the chance of recalling a payment falls with every hour. A practical first-hour plan covers calling the sending bank to request a recall, preserving the meeting invite, chat logs, any recording and the email thread, reporting to police and, in the US, to IC3, and notifying insurers. Preserved media can later be examined forensically to show what the attackers presented.

Where Deepfake Detection Fits After Arup

Detection is a control, not a replacement for payment verification. Its value is in giving staff and systems an objective signal at the moment someone is being persuaded, and in giving investigators evidence afterwards. Neural Defend's NeuroDetect handles the second job: an analyst uploads an authorised recording, screenshot or audio file into a case, and the analysis records model and version context, highlights the regions, frames and audio segments behind the result with explainability heatmaps, and produces a forensic report with observations, uncertainty and provenance for legal, insurance or law enforcement use.

For voice-led impersonation, NeuroCall (Beta) analyses calls for synthetic speech. For identity checks where a face is presented, NeuroVerify runs six signals in one API call and returns a unified risk score with signal-level evidence in under 200 ms p95 on eligible image requests. In each case the score feeds a policy, such as forcing a call-back, rather than deciding on its own.