Welcome back, my aspiring OSINT investigators!
In OSINT investigations, we often run across a face that does not appear anywhere else on the Internet. When we run image searches on the major image search engines such as Google, Bing, Tin Eye or even the facial recognition software, PimEyes, we come up empty-handed. This could be because of one of two reasons;
The image is unique and has never before appeared on the Internet;
The image is an AI generated deep fake.
Increasingly, I am finding it's the latter.
What are Deep Fakes?
Deep fakes are artificial intelligence (AI) generated video or images. Most often, they are a familiar face on another body (especially in porn) or a familiar face saying things they never said. These deep fakes are synthetic faces that are generated by AI. The woman below does not exist. She was generated by AI.
These deep fakes are becoming increasingly troublesome as online videos are appearing of political figures saying things they never said, porn videos of people who never did porn (most often women) and identities appearing on LinkedIn or dating sites of people who don't exist. When these deep fakes appear, they can fool voters, ruin marriages and careers and assist scammers to take your money.
For an example of what can be done with deep fake technology, check out this video of former US President Barack Obama.
How are Deep Fakes are Made
For many years now, movie producers have been using computer technology to create things on video that do not exist in reality. That's great; their job is to create fantasies. Unfortunately, this same technology is available to an increasing number of practitioners at lower cost and it enables them to use it for all sorts of purposes.
In face swap technology, the user runs thousands of face shots of two people through an AI encoder, say Face A and Face B. This algorithm then finds where the two faces are similar and compresses them. A decoder then can recover the Face A where Face B had been. The decoder reconstructs the face of Face A with the facial expressions and movements of Face B.
Where we have artificial images such as the one at the top of his post, GAN or generational adversarial network is used. The generator is fed random data and turns it into an image. This image is then fed a stream of data of real images. Initially, the image does not look like a person, but when repeated many, many times, it is possible to generate a face that does not exist in reality.
The Harm
These deep fakes, like any technology, can be used for good or bad. This technology has been unfortunately used to create pornography with the face of a celebrity or an ex-girl (boy) friend. Some of these have damaged careers and relationships. In addition, synthetic faces that do not exist in reality are used on social media and online dating sites by scammers. These synthetic people can be compelling and people give up their money easily to these synthetic lovers. Millions of dollars have been lost.
The Solution
Although these deep fake technologies are getting better and better, they still leave some telltale signs. In most cases, these "signatures" of fake can be detected and documented. If you have been a victim of deep fake technology--videos or still--we can probably help you. By analyzing the image(s), we can determine --and maybe more importantly--certify that the images are fake or not. We can provide you with a certified document that says these videos are (only if they actually are) fake. This can often be valuable to spouses, dates, partners, employers and future employers. For more information, click.