250 Million Dollars Disappeared with a Fake Video Call

In January 2024, an incident occurred at a company in Hong Kong. A video call came in from the CFO. On the screen was a familiar face and a well-known voice. A colleague was also present beside him. “We need to move funds urgently” — the amount transferred by the staff member who followed the instructions was $25 million (approximately 3.7 billion yen).

It was all fake. The CFO’s face and voice, as well as the colleague’s presence, were all generated by AI deepfake technology.

This is a story about a large corporation. But can small and medium-sized enterprises (SMEs) consider it someone else’s problem? Quite the opposite. Large corporations have multi-layered approval processes. They have security teams. SMEs do not have such structures. Because they operate on the CEO’s command, when the CEO’s face and voice are forged, there is no mechanism to stop it.

And now, Google has made that forgery even easier.

Google Introduces a Feature That Allows Anyone to Create Deepfakes

Google has introduced a new feature in YouTube Shorts that allows users to easily create and insert their own digital avatars into videos. By simply capturing a face, AI generates a realistic avatar. The expressions and movements are natural.

Google’s intention is likely to broaden the creative expression of creators. However, technology is not always used as intended. If this feature is misused, fake videos using the faces and voices of specific individuals could be created without any specialized knowledge.

The essence of the problem is not that “deepfakes can be created.” It is that “deepfakes cannot be detected.”

According to recent research, the accuracy of current deepfake detection technology is approximately 55.5%. This is almost the same probability as a coin toss. In other words, AI can only detect half of the fake videos it creates. It is even more difficult for the human eye to discern.

“Synthetic Trust Attacks” — The Structure of a New Scam

This new method of fraud is called “Synthetic Trust Attacks.” The difference from traditional phishing scams is clear.

Traditional phishing: A suspicious email arrives → The wording is strange → You notice it.
Synthetic Trust Attack: A familiar face speaks on a video call → The voice and expressions are almost indistinguishable from the real thing → You do not notice.

The cost of attacks has also dramatically decreased. Previously, creating advanced deepfakes required specialized skills and expensive GPUs. Now, with publicly available tools and cloud GPUs, it is possible to create sufficiently high-quality deepfakes in just a few hours. The cost is only a few thousand yen.

If the CEO of an SME appears on social media or YouTube, that alone provides enough training data. If there is a greeting video from the CEO on the company’s website, a voice clone is also possible. For attackers, the CEO of an SME is an ideal target because they have “readily available material and weak defenses.”

Three Defense Lines SMEs Should Implement Immediately

It is impossible to completely prevent this with technology alone. In a world where detection accuracy is 55.5%, relying solely on technology is dangerous. What is needed is a three-layer defense of technology + operations + culture.

Defense Line 1: Use Electronic Signatures to Prove “Genuine Instructions”

Implement electronic signatures for important documents such as purchase orders, contracts, and fund transfer instructions. Services like CloudSign or DocuSign can be used for a monthly fee of just a few thousand yen.

The key point is to formalize a rule that states, “Do not transfer funds based on verbal instructions via video calls or phone calls.” No matter how realistic the video is of the CEO saying, “Transfer the money immediately,” nothing will be done without an officially signed document. Just having this rule in place can neutralize most synthetic trust attacks.

The cost is under 10,000 yen per month, and the implementation period is one week. If this can prevent a loss of 3.7 billion yen, the cost-effectiveness is immeasurable.

Defense Line 2: Human Authentication via a “Password”

An analog system, rather than technology, can surprisingly be effective. Establish a “password” within the company.

Specifically, during high-value transfers or important decision-making, confirm a pre-agreed phrase. For example, ask, “What is this month’s keyword?” If the correct answer is not provided, it can be judged as fake. The password should be changed once a month.

This may be low-tech, but it cannot be bypassed by deepfakes. AI can replicate faces and voices, but it cannot replicate information known only to that person. The cost is zero, and it can be started today.

Defense Line 3: Utilizing AI Watermarking

In the medium to long term, embedding AI watermarks (digital watermarks) in the content your company produces can also be effective. Recent research has developed a framework called “Gaussian Shannon” that embeds invisible information in the early stages of the generation process.

This will allow for technical verification of whether “this video is real or fake.” By adding watermarks to your official videos or messages from the CEO, it will be possible to distinguish them from forged videos.

Although still in the early stages of adoption, there is a high likelihood that practical services will emerge in the next one to two years. It is advisable to keep an eye on the developments of this technology.

An Era Where the Cost of “Trust” Changes

The evolution of deepfake technology fundamentally changes the cost structure of “trust.”

Until now, the trust of SMEs has been built on the visibility of the CEO’s face and the ability to speak directly. This is especially true for local companies. The face and voice served as collateral for trust.

Now, that face and voice can be forged for just a few thousand yen. This means that simply having a visible face is no longer sufficient collateral for trust. To protect trust, a system is needed.

The introduction of electronic signatures, the operation of passwords, and the use of watermarks are all not extravagant investments. They can all be started for under 10,000 yen per month and within a week of implementation.

However, the fate of a company can be determined by whether it has implemented these measures or not in the event of an attack. A loss of 3.7 billion yen can be endured by a large corporation, but for an SME, even a loss of a few million yen can be fatal.

“We are too small to be targeted” — this misconception is the biggest security hole. Because they are small, their defenses are weak. Because their defenses are weak, they are targeted.

Let’s start with what can be done today. First, create a rule that “funds will not be transferred based solely on verbal instructions.” Just that will significantly protect your company’s trust.