The CEO’s Voice Can Be Stolen for $500. Is Your Company Safe?

Let’s start with the facts.

In June 2025, the U.S. Department of Justice arrested three members of a fraud group using AI deepfakes. The total amount defrauded exceeded $100 million (approximately 15 billion yen). The method was simple: they forged the voices and faces of corporate CEOs and legal advisors using AI, instructing employees via video calls or audio to “urgently transfer funds to this account.” Believing they were communicating with the real CEO, employees executed transfers amounting to millions of dollars.

Did you think this was only a problem for large corporations?

Unfortunately, small businesses are at even greater risk. The reason is clear: the cost of generating AI voice clones that mimic a CEO’s voice has now dropped to below $500. A few seconds of material from YouTube greeting videos, presentation audio from exhibitions, or phone recordings is sufficient. While large corporations have security departments, small businesses often have a culture of “acting without doubt when the CEO calls.” For attackers, there are few targets as cost-effective as this.

The Full Picture of the Incident: What Happened?

The arrested individuals were Jacob Saul, Sharon Hazan, and Guy Sheetrit. According to the U.S. Department of Justice, they systematically committed crimes from 2024 to 2026.

Breaking down their method:

1. Collect CEO Information from Target Companies: Acquire audio and video data from social media, investor relations videos, and speeches.
2. Duplicate the CEO’s Voice and Face Using AI: Combine voice cloning tools with deepfake video generation.
3. Contact Employees as the “Real CEO”: Directly instruct them via video conference or phone, using scenarios like “urgent funds needed for a confidential M&A deal” or “confirmed by the lawyer.”
4. Execute the Transfer: Have them transfer funds to overseas accounts, immediately dispersing and laundering the money.

The key point is that employees believed they were “executing their duties correctly” rather than being “deceived.” With the CEO’s face visible and instructions coming in their voice, there was no reason to doubt. This is the decisive difference from traditional phishing emails.

What the Collapse of Costs Means

I want to discuss this from a structural perspective.

In the past, perfectly mimicking someone else’s voice required the skills of voice actors or impersonators. Creating fake videos required Hollywood-level VFX teams, with costs in the tens of millions of yen. Therefore, fraudsters had no choice but to deceive through “email content.”

Now, the cost of voice cloning is in the tens of thousands of yen. Real-time video synthesis is possible with subscription tools costing just a few thousand yen per month. In other words, the costs of “identity verification” and “identity forgery” have reversed.

This represents a structural change.

Traditional defenses relied on verifying whether someone was who they claimed to be through sight and sound. The voice on the phone, the face in the video—this was the foundation of trust. However, AI has almost zero-costly destroyed that foundation. Our eyes and ears can no longer be trusted.

I want to ask small business owners: If a transfer instruction comes from the CEO via phone, how many employees would execute it without doubt? Probably almost all of them. That is the greatest risk.

The Democratization of Attack Tools: Another Threat

In parallel with this incident, there is another movement to watch: the “democratization” of cyber attack tools.

According to reports from security researchers, a hacking framework called Artex is publicly available on hundreds of servers. The attack capabilities that were once only available to state-level hackers are now circulating like open source.

Additionally, there are analyses indicating that AI models from China (like DeepSeek and GLM) are rapidly enhancing the code generation capabilities for cyber attacks. Automatic generation of attack codes and automatic exploration of vulnerabilities—this is becoming an era where even “unskilled attackers” can use these tools.

In short, the costs and barriers to entry for attackers have dramatically decreased. On the other hand, defenses—especially for small businesses—have hardly changed in the past five years. This asymmetry is the essence of this incident.

Four Things Small Businesses Should Do “This Week”

Abstract calls to “raise security awareness” are meaningless. Here are four specific actions that can be taken this week.

1. Include a “Code Word” in Transfer Rules (Cost: $0, Time Required: 30 minutes)

Do not execute transfer instructions over $10,000 solely based on phone or video calls. Establish a rule to confirm a pre-decided code word. Change the code word monthly. While AI can mimic the CEO’s voice, it cannot steal a code word known only to the CEO and the finance officer.

This alone can almost nullify the methods used in this incident.

2. Mandate “Callback Confirmation” (Cost: $0, Time Required: 15 minutes)

If there is a transfer instruction from the CEO or executives, create a rule to always call back to confirm. Attackers can “call in,” but they cannot “answer a callback.” Simply calling back the CEO’s mobile number will dismantle the impersonation.

3. Inventory the CEO’s Audio and Video Material (Cost: $0, Time Required: 1 hour)

Greeting videos uploaded to YouTube, webinar recordings, audio posts on social media—all of these can serve as “voice material.” Make unnecessary items private. While there is no need to delete them completely, raising the cost of material collection for attackers can still be effective.

4. Communicate to All Employees to Doubt Instructions from the CEO’s Voice (Cost: $0, Time Required: 5 minutes in a Morning Meeting)

Simply share the overview of this U.S. incident during a morning meeting. “We are in an era where the CEO’s voice and face can be faked by AI. Therefore, even if you receive transfer instructions via phone or video, always confirm with a callback”—say this from the CEO’s own mouth. The CEO stating “Doubt my voice” is the greatest defense.

All of these actions cost $0. While implementing security software and external audits are certainly effective, those can wait until next month. The four actions above can be done this week.

The Need for Redesigning “Trust”

Let’s discuss a bit further into the future.

What this incident indicates is that the trust mechanism that humanity has relied on for tens of thousands of years—”verifying identity through voice and face”—is beginning to break down.

So, what will we rely on to ensure trust? The answer is a combination of knowledge authentication and procedural authentication. In other words, a double-check of “information known only to the individual” and “whether the pre-decided procedure has been followed.” This same principle of two-factor authentication used by banks can be applied to internal communications.

Large corporations solve this through systems: approval workflows, electronic signatures, zero-trust architecture. However, such investments are not realistic for small businesses.

That’s why small businesses can win through “human operation.” Code words, callback confirmations, sharing in morning meetings. They can establish defenses through “rules and habits” rather than technology. This is something large corporations cannot replicate. Because while it is difficult for large corporations to enforce the same rules across 10,000 employees, a company of 10 can communicate it to everyone in one morning meeting.

The “smallness” of small businesses can actually become a weapon in terms of security.

What to Watch Going Forward

There are three key points.

1. Improvement in Real-Time Deepfake Accuracy: Currently, the level is such that it is “unnatural if you look closely,” but in six months, it is highly likely that it will become indistinguishable. The defense strategy of “verifying by sight” is nearing its expiration date.

1. Improvement in Japanese Voice Cloning Accuracy: In English-speaking regions, it is already at a practical level. While Japanese still retains some unnaturalness, it is only a matter of time. If Japanese small businesses think, “We are safe because we speak Japanese,” that is a dangerously optimistic view.

1. Automation of Attacks: Currently, humans manually select targets and create scenarios. If this becomes automated with AI, the reality of “scattershot CEO scams” targeting thousands of companies simultaneously will emerge. If that happens, small businesses will not be “not targeted” but will instead become automatic targets.

Conclusion: Business Decisions in an Era Where the CEO’s Voice Can Be Stolen for $500

The lesson from this incident is simple.

The “CEO’s voice and face” no longer function as means of identity verification.

And the first step in defending against this threat is not the purchase of security products, but the CEO personally telling all employees, “Doubt my voice.” Cost: $0. Time required: 5 minutes.

Technological evolution cannot be stopped. However, we can “rebuild trust broken by technology” through “operations.” And smaller organizations can do this much more quickly.

To small business owners: I urge you to discuss this in your morning meeting today. This will be the first step in protecting your company from a $150 million fraud.