The Twitter and social media sphere went bonkers recently as @deeptomcruise posted some seriously good deepfakes. So far, it has been pretty easy to spot deepfakes. Not so in these videos. Just a few small glitches that could easily fly under the radar. A small voice difference and a small mouth sync in one frame. The creator managed to eliminate most of the usual red flags. No obvious blurred edges around the mouth, no restricted or limited facial and body movements. Deepfake Tom did a variety of moves in . These were scary good. You can easily see how trusting your own eyes may not be good enough to determine if a circulating video is real or not without special AI/machine language software looking for machine learning artifacts. This is scary bad news because it shows how quickly GAN (generative adversarial network) technology is evolving. Since deepfakes now usually require a few real video samples to learn from you might be tipped off if you’ve seen the real videos. But otherwise you may think its the real thing.
Can I trust My Eyes and Ears?
The future scenarios for social engineering are unsettling. Imagine how nation states might use this to erode trust and cause political instability around the world. Social engineering scammers will undoubtedly use deepfakes to up their game in BEC scams and Invoice scams. Imagine the scammer calling an employee who typically manages financial transfers. Suppose they are trained to verify such calls and call the CEO or COO or a third party supplier. The scammer sets up a call time and spoofs the telephone number and then uses a segment of deepfake video to try to get around zero trust and verify. Employees will always need to be trained to use multiple verification methods as scammers will also use AI and machine learning as an adversarial tactic.
Input Magazine has the background story on @deeptomcruise. Highly recommended to check out this video since its been getting lots of views and share. It’s now a hot story on many social media platforms and security media. Let me know what you think.