Written by: Mio Kanzaki(Security Analyst (AI))Reviewed by: 千葉圭介Updated
A deepfake is AI-made video or audio that looks just like the real thing. NIST's material for generative AI describes it as synthesized video, audio and photorealistic images that look real to the human eye.
How does a deepfake work?
It is a situation in which an AI takes samples of a real person's photos and voice and makes a fake that imitates that person's face and way of speaking. Like a skilled impressionist, it reproduces the features of appearance and voice with high accuracy.
NIST says that AI able to handle several kinds of information has made it possible to create extremely realistic deepfakes in addition to false information in text. Such synthetic content becomes something that tests the ability to tell what people made from what AI made.
As one way of dealing with it, NIST lists mechanisms that trace the origin and history of content (provenance records) and the detection of synthetic content. It explains that these help to give users material for judging whether something is genuine.
The whole picture of fraud is gathered in phishing measures. To see what your own site shows from outside, use the free URL check.
Why does a deepfake matter?
Voice and face have long been used as clues for confirming who someone is. If they can be forged, there is danger in acting on a request made by phone or video meeting as it is.
The AI Guidelines for Business (Ministry of Internal Affairs and Communications and Ministry of Economy, Trade and Industry) point out that generative AI has made it possible for anyone to create information that poses as true or fair in content. They ask you to recognize that the danger of false information disrupting society is rising.
Even a small company can prepare. For requests such as sending money, changing a password or sending important documents, do not decide by voice or video alone, and set a rule of confirming with the person by another means. An example of a request posing as an executive is explained under business email compromise, and tricks by phone under vishing.
For help with building the procedure, contact us.
Related terms
- Business email compromise (BEC): fraud that poses as a business partner or executive and gets money sent.
- Vishing: a trick that deceives the other party by using the telephone.
- Social engineering: tricks that draw out information by exploiting gaps in a person's psychology and behavior.
- Generative AI: AI that newly creates text, images and the like.
Next step
Decide the procedure for confirming who someone is when you are asked to send money or perform an important operation. Make it a rule not to judge by voice or video alone, and to confirm through another contact channel. The whole picture of measures is in phishing measures. What is visible from outside can be seen with the free URL check, and we take questions on building the procedure through a free consultation.
Frequently asked questions
Can a deepfake be spotted at once just by looking?
Sometimes it is hard to tell. NIST says that the output of generative AI such as deepfakes tests the ability to tell content made by people from synthetic content made by AI. You need a way of working in which you do not decide something is real from the impression of the look or the voice alone.
Where do we start to prepare for deepfakes?
For requests involving money or information, do not judge by voice or video alone. Create a procedure of confirming through a separate contact channel decided in advance. NIST says that mechanisms that trace where content came from, and the detection of synthetic content, can help in judging what is genuine.
Frequently asked questions
Can a deepfake be spotted at once just by looking?
Sometimes it is hard to tell. NIST says that the output of generative AI such as deepfakes tests the ability to tell content made by people from synthetic content made by AI. You need a way of working in which you do not decide something is real from the impression of the look or the voice alone.
Where do we start to prepare for deepfakes?
For requests involving money or information, do not judge by voice or video alone. Create a procedure of confirming through a separate contact channel decided in advance. NIST says that mechanisms that trace where content came from, and the detection of synthetic content, can help in judging what is genuine.