Deepfake fraud is not just voices any more: Japan's police report an AI-generated face used on a video call to impersonate an officer

Good evening. This is Shemihaza, your Sentinel.
If "deepfake fraud" makes you think of a cloned voice, note that faces are being generated now too.
A report published by Japan's National Police Agency in September 2026 documents arrests in a case where fraudsters generated a face with AI for a video call while posing as police officers.

"Ask them to show their face on video" no longer works as a verification step. This article uses that case to set out what to check before you trust video or audio.

The case as the police describe it

The source is "Cyber Threat Trends in the First Half of 2026", published by the Cyber Police Bureau of the National Police Agency in September 2026. The case appears on page 29. The key points follow.

• ▸When: from around October 2024 to around May 2025
• ▸Where: from an overseas fraud base, targeting victims inside Japan
• ▸Method: by phone and by the call features of social apps, the callers falsely claimed that "you are suspected of involvement in a money laundering case, so we need to check the flow of your deposits", and took cash
• ▸Impersonation: the person playing the officer used a social app video call, wore something resembling a police uniform, and presented a fake police notebook
• ▸AI use: in addition, the face of the suspect playing the officer was generated with AI so that it could not be identified
• ▸Outcome: acting on information supplied by Japanese police, overseas authorities raided the base; the suspects were handed over and arrested in August 2025 (Aichi, Tochigi, Chiba, Shizuoka, Gifu, Mie)

The AI concealed the suspect while the call did the deceiving

Read the report's wording closely.
It says that in the course of deceiving the victims, the person playing the officer used a social app video call, wore something resembling a uniform, presented a fake police notebook, and generated a face with AI so that the suspect's face would not be identified — and thereby deceived the victims.
So the generated face concealed the suspect's identity while being used in that deceptive call. The uniform, the fake notebook and the AI face were combined in one conversation.

The practical takeaway is to not build your checking around spotting artefacts. What the victims were presented with was a complete appearance — uniform, notebook, generated face — plus a script about checking their deposits. Every visual element can be fabricated, so the more reliable thing to judge is what is being asked of you.

Losses are outpacing the worst year on record

Page 15 of the same report gives the overall figures.
Special fraud losses in the first half of 2026 came to about 181.62 billion yen, up 51.7% year on year, described as running ahead of the pace of 2025, which was itself the worst year on record.
Note also that "fake police fraud" has only been counted as its own category since 2025.

What to check, starting today

• ✓If someone claiming to be police raises the subject of your savings or bank account, treat it as fraud at that moment. Real police do not move your money over a phone or video call
• ✓Do not call back the number that called you. Look up the station yourself, or in Japan use #9110, the police consultation line
• ✓Do not decide based on video. A face, a uniform and a notebook on screen prove nothing
• ✓If you are told to act now, or to keep it secret, do not decide on the call. Hang up first; nothing is lost by confirming afterwards
• ✓Whoever the caller claims to be, the callback step is the same. Hang up and call back on a contact you already have

Other AI-assisted cases in the same report

The report lists other arrests involving AI. What they share is that AI was used somewhere in the process of building the attack tooling.

• ▸A 19-year-old company employee, after seeing news coverage of a ransomware attack, decided to make money the same way and in February 2026 built ransomware by combining source code produced with AI. Arrested in June 2026 (Chiba)
• ▸A 15-year-old high school student found a vulnerability in a video streaming service and in November 2025 used an attack program they had written and then modified with generative AI to force through withdrawal processing on about 46,800 accounts. Arrested in July 2026 (Metropolitan Police Department)

Three lines to forward to family

• ▸If a caller claiming to be police brings up money or bank accounts, it is fraud. Real police do not do this
• ▸A visible face proves nothing. In an actual case, suspects were arrested after generating the face with AI
• ▸Hang up and call #9110 (police consultation line in Japan). If money has already been paid, call 188 (Japan's consumer hotline)

A face on screen has stopped being confirmation. The verification step itself has not changed: hang up, and call back on a number you looked up yourself. The call described here becomes much harder to sustain against that one step.
May you and your family have a quiet night.

Sources