AI Cybersecurity, Explained. Plain-language security for everyone, with practical steps for the organizations that answer for AI at work.
In early 2024, a finance employee in the Hong Kong office of Arup, the British engineering firm behind the Sydney Opera House, received a message that appeared to come from the company’s UK office. He suspected it was a phishing email. Then he joined a video call. On screen were people he believed were the CFO and other colleagues. They looked and sounded right. According to CNN, citing Hong Kong police and the public broadcaster RTHK, he went on to send HK$200 million, about US$25.6 million, in 15 transactions. Every other person on that call was a deepfake (Magramo, 2024).
Arup confirmed to CNN that “fake voices and images were used,” and said its financial stability and internal systems were not affected (Magramo, 2024). The case has since been cited by the US Treasury’s financial crimes unit in a formal alert to banks (FinCEN, 2024, p. 5). It is the best-known example of a fraud that is no longer rare, and it carries a lesson that does not depend on spotting fakes at all.
How the scam works
Deepfake fraud is an old trick with a new costume. The trick is impersonation: pretend to be someone the victim trusts, create urgency, and ask for money, passwords or codes. What artificial intelligence changes is how convincing and how cheap the costume has become.
In December 2024, the FBI warned that criminals use generative AI to make fraud more believable and to commit it at larger scale, noting that these tools can even correct the small errors that used to give scams away (FBI, 2024). The FBI listed the patterns it sees: cloned voices of relatives claiming to be in a crisis and needing money; cloned voices used to get into bank accounts; and fake real-time video calls with people posing as company executives, police or other officials (FBI, 2024).
The same technique is aimed at people in public life. Since at least 2023, according to the FBI, impersonators have sent texts and AI-generated voice messages posing as senior US officials, including White House and Cabinet-level figures and members of Congress, then moved targets onto encrypted messaging apps to ask for money, passport copies or authentication codes (FBI, 2025a, 2025b). The FBI’s advice is blunt: if a message claims to come from a senior official, “do not assume it is authentic.” It also warns that AI voice cloning “can sound nearly identical” to a real contact (FBI, 2025b).
The numbers are growing. In its 2025 annual report, the FBI’s Internet Crime Complaint Center said it received more than 22,000 complaints containing AI-related information, with reported losses above $893 million. Those included more than $30 million lost to business email compromise scams involving AI, and more than $5 million lost to “distress” scams in which voice cloning mimics a loved one in trouble (FBI IC3, 2026, p. 39).
Key idea: out-of-band verification
Out-of-band verification means checking a request through a separate channel that you choose, not the one the request arrived on. If your “boss” asks for a payment on a video call, you hang up and ring them on the number you already have. It matters because it works even when the fake is perfect: a scammer can copy a face and a voice, but cannot easily answer the phone number you already know belongs to the real person.
Why “spot the fake” is not enough
Most advice about deepfakes focuses on looking for flaws: odd hands, strange shadows, a lag between voice and lips. The FBI does recommend watching and listening carefully (FBI, 2024). But notice what happened at Arup. The employee was suspicious at first, and the video call is what reassured him. A good enough fake turns the very act of checking into proof.
That is why the most reliable defences do not depend on the quality of the fake. The FBI’s tips centre on process: hang up and call the person or organization back on a number you already know; agree a secret word or phrase with family members; never share one-time security codes; and do not send money to someone you have only met online or by phone (FBI, 2024, 2025a).
Banks are told to watch for behaviour, not just pixels. The Treasury’s Financial Crimes Enforcement Network (FinCEN) reported an increase since 2023 in suspicious activity reports describing suspected deepfakes in fraud (FinCEN, 2024, p. 1). Among its red flags: a person who, during a live identity check, tries to switch to a different communication method because of “excessive or suspicious technological glitches” (FinCEN, 2024, p. 6). A sudden bad connection at exactly the moment you try to verify someone is itself information.
How organizations stop it
For a business, the answer is to make large payments impossible to authorize on the strength of a single conversation, however convincing. In practice that means:
- A callback rule for any new payment instruction or change of bank details, using contact details already on file, never ones supplied in the request.
- Two people for payments above a set amount, so one persuaded employee is not enough.
- Permission to say “I will call you back.” Staff must know that verifying a senior person’s request will never be held against them, and that secrecy or urgency is a reason to slow down.
- A record of how each high-value request was verified, so the check is a real step, not a memory.
These are approval gates: points where a high-impact action needs a verified human yes. They work against a deepfake for the same reason they work against an ordinary con artist.
What the record does not show
Much of the detail about the Arup case comes from police statements reported in the press; Arup said it could not discuss details while the investigation continued, and the reporting we read did not mention any arrests in the case (Magramo, 2024). We do not know exactly which tools the fraudsters used. The FBI figures count complaints that mention AI, which depends on victims recognizing and reporting it; the true scale is likely different, in either direction. And none of these sources tell us how often deepfake attempts fail, which is the number that would show how well verification habits work.
What to do with this
- At home: agree a family code word today, and if a loved one calls in a crisis asking for money, hang up and call them back on their usual number.
- At work: if a senior person asks you to move money or share codes urgently or secretly, verify through a channel you choose, whatever you see or hear.
- If you run payments: check that no single call, email or video meeting can release a large payment without an independent callback and a second approver.
- If you lead an organization: say out loud, in writing, that verifying your requests is expected, not insubordinate.
Go further. Operational AI Readiness is FairByDesign’s free field manual for AI that is already live. Chapter 9, Approval Gates for High-Impact Actions, shows how to design a human yes that cannot be skipped under pressure, and Chapter 5 explains why oversight only works when people are empowered to say no. Read it free.
Your next step: scams change faster than advice pages. Get the free Mercury Brief and we will tell you, in plain language, when a new pattern appears and what still works against it.
FairByDesign is reader-funded. If this helped, get the free Mercury Brief or see ways to support. Support never buys editorial control.
References
Bakos, V. (2026). Operational AI readiness: A practitioner guide to governing AI that is already live. FairByDesign. https://fairbydesign.org/operational-ai-readiness/
Federal Bureau of Investigation. (2024, December 3). Criminals use generative artificial intelligence to facilitate financial fraud (Alert No. I-120324-PSA). Internet Crime Complaint Center. https://www.ic3.gov/PSA/2024/PSA241203
Federal Bureau of Investigation. (2025a, May 15). Senior US officials impersonated in malicious messaging campaign (Alert No. I-051525-PSA). Internet Crime Complaint Center. https://www.ic3.gov/PSA/2025/PSA250515
Federal Bureau of Investigation. (2025b, December 19). Senior U.S. officials continue to be impersonated in malicious messaging campaign (Alert No. I-121925-PSA). Internet Crime Complaint Center. https://www.ic3.gov/PSA/2025/PSA251219
Federal Bureau of Investigation Internet Crime Complaint Center. (2026). 2025 IC3 annual report. https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf
Financial Crimes Enforcement Network. (2024, November 13). FinCEN alert on fraud schemes involving deepfake media targeting financial institutions (FIN-2024-Alert004). U.S. Department of the Treasury. https://www.fincen.gov/sites/default/files/shared/FinCEN-Alert-DeepFakes-Alert508FINAL.pdf
Magramo, K. (2024, May 16). British engineering giant Arup revealed as $25 million deepfake scam victim. CNN (via ABC 17 News). https://abc17news.com/money/cnn-business-consumer/2024/05/16/british-engineering-giant-arup-revealed-as-25-million-deepfake-scam-victim/


Leave a Reply
You must be logged in to post a comment.