Japan report outlines AI voice-cloning fraud targeting corporate accounting staff
ITmedia Enterprise reported on Sept. 3 that an AI-driven fraud method involving cloned executive voices is already being used in Japan. The report, which cited a former Tokyo Metropolitan Police Department official, put the damage figure at JPY 4.5 billion. It did not say whether that number referred to one case or a cumulative total, and it did not specify the period covered.
The method itself is straightforward. Fraudsters use AI to clone a company president’s voice, then call the person in charge of accounting and instruct them to send money.
The script described in the report goes like this: 「Transfer JPY 10 million to the client today. This is a confidential M&A deal, so do not tell anyone.」 The call combines urgency with secrecy, making it less likely that the accounting staff will try to verify the request with the real executive.
Three conditions were highlighted as key to the scam working
The report broke the success factors into three parts: the incoming call appears to come from the president’s real mobile number, the voice sounds like the real person, and the caller keeps pressing that the matter is urgent. Put together, those elements make the payment request appear credible.
Publicly available executive audio can become training material
According to the report, the barrier to voice cloning has dropped sharply. Only tens of seconds of voice samples may be enough to generate a cloned voice with practical quality. A short self-introduction or a brief opening remark can already provide enough material.
Collecting the sample may be easier than the technical process itself. The voices of company presidents and senior executives can often be found in earnings briefings, recruiting videos, seminar recordings, and YouTube uploads. Those materials are often published by the companies themselves. The more a company emphasizes external communications, the more voice material may be available for cloning.
Suggested defense: verify transfer requests through a separate channel
The article’s recommended response is simple: any payment instruction delivered by phone should be checked through another channel, such as calling back a known company number or confirming the request through an internal system.
As described in the report, the JPY 4.5 billion case depended on three things being present at the same time: caller ID, a convincing voice match, and repeated pressure to act quickly. It also said a usable clone may be built from just tens of seconds of audio, without any direct contact with the target, because samples can be pulled from public appearances and online videos.
Two other AI-related fraud cases were also mentioned
The original article also said Japan saw two other AI fraud cases of a different nature in the same week, though no further details were included in the provided summary.