PROBLEM: Deepfake scams are spreading fast as cloned voices and fake video calls impersonating executives leads to U.S. losses heading toward tens of billions of dollars a year.
SOLUTION: Japan’s National Institute of Informatics (NII), a Tokyo research institute, has built a deepfake detector, Synthetiq Vision, now being folded into a national platform that flags manipulated audios, images, and videos.
Deepfake fraud moves from novelty to industry
Fraud that hijacks a familiar face or voice has moved from novelty to industry. In early 2024, a finance worker at the engineering firm Arup was tricked into sending $25 million to criminals after a video call in which every colleague, including the chief financial officer, was AI-generated. According to Japan’s National Institute of Informatics, such attacks are cheap as a convincing voice clone needs only a few seconds of audio.
The scale is climbing. Deloitte’s Center for Financial Services projects U.S. generative-AI fraud losses rising from $12.3 billion in 2023 to $40 billion by 2027. In 2025, the FBI’s annual cybercrime report logged for the first time, AI-related complaints as their own category, with roughly 900 million in losses.
A Japanese institute builds a possible countermeasure
NII has set up a Research Center for Synthetic Media in 2021 and has licensed its detector, Synthetiq Vision, since 2023. Synthetiq Vision traces back to a deepfake-video detector named “MesoNet” published in 2018. That tool is now one piece of a larger effort: since October 2024, a nine-organization consortium led by Fujitsu, including NEC Corporation, has been building what it calls a first-of-its-kind disinformation platform, with a prototype finished by mid-2025. NII and NEC handle the detection layer. It is being trialed first with fact-checkers and government offices, with company and newsroom use targeted from fiscal 2026, focused on election and disaster misinformation.
Expanding its horizons across AI content
Last month, NII has made a big update to the deepfake-detection tool. On top of its original focus on fake images and videos of people’s faces, it can now also spot fake “disaster” images like AI-made photos of floods, fires, and collapsed buildings that spread on social media and cause panic during real emergencies. The tool also got a technical upgrade so it runs on the cloud, letting it check huge numbers of images quickly and connect easily to news, security, and monitoring systems. With Japan being disaster prone, NII may have the upper-hand in understanding the intricacies between real and AI generated disasters videos, however little can be said with how much this affects the program’s accuracy. What makes Synthetiq Vision standout over many other AI detection softwares is their open source approach, as NII provides their program free to use.
Why detection alone won’t win
NII is not alone, Intel’s FakeCatcher, Reality Defender and Sensity are among global rivals in a detection market whose 2026 size estimates swing from under $1 billion to several billion, a sign of how new the field is. And detection has hard limits: on a 2024 benchmark of real-world fakes, established tools lost roughly 45 to 50 percent of their accuracy, because detectors trained on older fakes struggle against newer ones. Analysts call it necessary but not sufficient, urging firms to keep human checks such as second-channel verification and code words. NII’s researchers are more optimistic, arguing that models trained on real human data and across thousands of AI generators can start catching fakes they were never shown. Japan’s bet is that reliable detection becomes shared public infrastructure for the AI era and not just another product.
