When the call sounds like the CEO, but isn't.
Synthetic voice and video are now within budget for ordinary fraud crews. The question is no longer if, but which protocol fires first. Below: the case data that should change your finance controls, a six-step authentication protocol, and the regulatory framework you can lean on after the fact.
Arup Hong Kong, late January 2024 — fifteen wires, twenty-five million dollars, four minutes per transfer.
The finance employee in Arup's Hong Kong office initially received an email from an account claiming to be the firm's UK-based CFO, asking that several confidential transactions be deployed. The employee suspected phishing — the right instinct. He was reassured when he joined a video conference where the CFO and several colleagues appeared on camera, looked and sounded like themselves, and walked him through the transfers in real time.[1]
Every other participant on the call was an AI-generated deepfake. Over the course of the call he authorized fifteen separate wire transfers, totaling approximately USD 25 million, to five accounts controlled by the perpetrators.[2] Arup confirmed the incident publicly in May 2024. The firm's chief information officer, Rob Greig, framed it explicitly: “None of our systems were compromised and there was no data affected... this was technology-enhanced social engineering.”[3]
The relevant lesson is operational, not technical. Every Arup system was intact. The break point was the company's payment-approval workflow — specifically, that a video call could substitute for an out-of-band callback above the wire-transfer threshold.
Three more incidents that did not make the front page.
A UK-based managing director, insured through Euler Hermes, was talked into wiring €220,000 (≈ USD 243,000) to a Hungarian supplier after a phone call with what he believed was his German parent-company CEO — recognizable by “slight German accent” and the “melody” of the voice. Funds were laundered through Mexico. A second attempted call was rebuffed when the originating number was identified as Austrian, not German.
Wall Street Journal, August 2019. Sophos analysis
In July 2024 a Ferrari executive received WhatsApp messages from a number resembling but not matching CEO Benedetto Vigna's, then a phone call using a deepfake of Vigna's voice and accent. The attempt failed when the executive asked the caller to name the title of a book Vigna had personally recommended a few days earlier — an answer the synthetic system did not have.
Bloomberg, 26 July 2024. Bloomberg coverage
In April 2024 a LastPass employee received calls, texts, and a voicemail using an audio deepfake of CEO Karim Toubba — over WhatsApp, outside normal company communication channels. The employee flagged the channel choice and forced urgency as social-engineering hallmarks and reported to internal security; LastPass disclosed the attempt to share lessons.
LastPass blog, April 2024. LastPass disclosure
Six controls. Adopt the lot.
Each step below is operational, not technical, and addresses a specific failure mode visible in the incidents above. The protocol assumes that the synthetic media will eventually look and sound flawless.
Fig. 04 · Each control breaks a specific failure mode visible in the Arup, Ferrari, LastPass, and 2019 Euler Hermes cases above.
Federal guidance and rules now address synthetic media directly.
Contextualizing Deepfake Threats to Organizations (12 September 2023) — co-authored by NSA, FBI, and CISA. Provides defensive recommendations for media authentication, user awareness, and detection integration.
FCC Declaratory Ruling FCC 24-17 (8 February 2024) confirmed that AI-generated voices in calls fall within the TCPA's restrictions on “artificial or prerecorded voice” and require prior express consent. Effective immediately on adoption.
Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (IC3 PSA, 3 December 2024) — formalizes the threat model and advises individuals and organizations on indicators and reporting.
Operational artifacts you can adopt today.
Public statement scaffolds for executive impersonation, fraudulent video, and astroturfed campaigns.
Available on request — info@imadethisup.org
A finance/AP procedure that tolerates voice-clone attacks at the change-of-banking-details step.
Available on request — info@imadethisup.org
Single-page laminate of the protocol above. Print at desk size for finance and executive assistants.
Available on request — info@imadethisup.org
Ten verified sources for this page.
- 001FBI Internet Crime Complaint Center (2024). 2024 Internet Crime Report.
- 002FBI / IC3 (2024). PSA241203 — Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud. 3 December 2024.
- 003NSA / FBI / CISA (2023). Cybersecurity Information Sheet — Contextualizing Deepfake Threats to Organizations. 12 September 2023.
- 004Federal Communications Commission (2024). Declaratory Ruling FCC 24-17. 8 February 2024.
- 005CNN (2024). “Finance worker pays out $25 million after video call with deepfake 'chief financial officer'.” 4 February 2024.
- 006CNN (2024). “Arup revealed as victim of $25 million deepfake scam involving Hong Kong employee.” 16 May 2024.
- 007World Economic Forum (2025). “Cybercrime: Lessons learned from a $25m deepfake attack.” February 2025.
- 008Bloomberg (2024). “Ferrari Narrowly Dodges Deepfake Scam Simulating Deal-Hungry CEO.” 26 July 2024.
- 009LastPass (2024). “Attempted Audio Deepfake Call Targets LastPass Employee.” April 2024.
- 010Wall Street Journal / Sophos (2019). Voice-clone CEO fraud at Euler Hermes-insured UK firm — $243K transfer.
War Room FAQ.
How big is the BEC problem?
The FBI's IC3 2024 Annual Report counted 21,442 reported business email compromise incidents in 2024 with $2.77 billion in adjusted losses. Total cybercrime losses across all categories reached $16.6 billion, a 33% year-over-year increase.
What was the Arup deepfake fraud?
In late January / early February 2024, an Arup Hong Kong finance employee was tricked into making 15 wire transfers totaling roughly USD 25 million after a video conference in which every other participant — including the supposed CFO — was an AI-generated deepfake. Arup confirmed the incident publicly in May 2024. CIO Rob Greig described it as “technology-enhanced social engineering.” (CNN, May 2024)
What is the single best protection against voice-clone fraud?
An out-of-band callback to a number stored in your corporate directory — never the number that initiated the request. Combine with a 30-minute soft hold on payments above a defined threshold and a pre-shared challenge phrase rotated quarterly. The IC3 Recovery Asset Team reports a 66% success rate freezing funds when reported quickly.
Are AI-voiced robocalls illegal?
In the United States, yes. The FCC's Declaratory Ruling FCC 24-17 (8 February 2024) confirmed that AI-generated voice calls fall within the TCPA's restrictions on “artificial or prerecorded voice” calls and require prior express consent. The ruling was effective immediately.
Where do I report a synthetic-media fraud?
File with the FBI Internet Crime Complaint Center at ic3.gov within 24 hours, preserve audio and message logs, and notify your bank's fraud team to attempt a recall. The IC3 Recovery Asset Team can attempt to freeze fraudulent transfers if alerted promptly.
Can our internal video-conferencing platform detect deepfakes?
Most consumer-grade conferencing platforms do not. Real-time deepfake detection is an open research area — challenge-response approaches (asking the caller to perform unexpected gestures or answer personal questions) currently outperform passive detectors in production. The Ferrari attempt was foiled informally by exactly this technique.