In this talk, the speaker outlines the rise of deepfakes—AI‑generated images, video, and eventually real‑time audio‑visual media that can convincingly impersonate individuals, especially public figures. He cites a recent Pennsylvania case where a mother allegedly used AI‑altered videos and photos to sabotage rivals in her daughter’s cheerleading squad, illustrating how even non‑criminal actors can weaponize the technology. The presentation highlights the emerging arms race between creators of synthetic media and detectors, noting a University of Buffalo tool that leverages corneal reflections to spot deepfakes with about 94% accuracy, while acknowledging that generative models will likely adapt. The speaker warns that once real‑time, smartphone‑accessible deepfake generation (combining video and audio) becomes feasible, cybercriminals and nation‑states will exploit it at scale, eroding trust in digital communications, stalling innovation, investment, and adoption. He stresses that the malicious potential is bounded only by attacker creativity and greed, urging vigilance.
Chapters
Let's spend a couple of minutes today talking about deepfakes and how they're really going to be the digital future of forgery. Now, we've seen in the news many deepfakes, and these are AI-generated personas, in many cases that of celebrities or, or somebody in the limelight like politicians. And the AI system enables an actor to basically take over the facial features and say whatever they want to say, and it looks like that person. And we're seeing all sorts of different, um, AI systems that can create very authentic-looking images of people, as well as deepfakes for videos. And we're seeing those again, especially for celebrities. We've been talking about how potentially this could be misused in the future, and I still hold the position, as I have for, well, well over a year now, that the cybercriminals, when it reaches a certain point, they're going to jump into this and that's when it's going to become a real problem. We'll talk about what that crescendo moment is shortly. But we're also seeing some, some interesting things. So, for example, just the other day, a Pennsylvania woman, she ended up making— and she's accused, she got arrested and she's accused of making deepfakes. To kick the rivals off of her daughter's cheerleading squad. Yes, AI-altered videos and photos of these other teens to portray them as smoking and drinking and being naked, all to get her daughter a better position on the squad. Now, if a mom in Pennsylvania can do this, just imagine what cyber criminals and nation states can do with this type of technology. It is a little concerning. Now, there is tension in the system. It is not just all one-sided. In fact, technology is being developed to help identify deepfakes, both videos and images. For example, uh, recently the University of Buffalo created a deepfake spotting tool and it is about 94% correct. It actually uses the reflections in the cornea of images, right, to detect whether it's real or whether it was generated by an AI. Now, this is just one maneuver, right? It was one paper, one tool. And the reality is generative AI systems, well, they should be able to compensate for that pretty easily. But it does showcase how there are opposing forces to both drive realism and to be able to detect synthetic images and videos. So this is a never-ending battle that we're probably going to see played out again and again and again. And right now, deepfakes, well, they're a novelty, but there will be a moment when they will be used as a very powerful tool to impersonate others and to manipulate and abuse victims. So this is the moment that we need to be concerned with. And again, I've been talking about this for well over a year, a year, and it's really when the following pieces come together, right? First off, you need to have real-time video and audio married together as part of a deepfake, and it has to be available for use on smartphones. And we're still a ways away here. Smartphones right now aren't configured to be able to do that stuff in real time. Audio is kind of behind the video side of it. So, you know, but these things move very, very fast, especially in the AI world where the iterations of improvements can come extraordinarily quickly down, down the pipeline. So it's in the distance, we don't know how far, but when those conditions are met, then we're going to see the creativity of everyday fraudsters to start taking advantage, to manipulate, steal, misinform, uh, people to a huge effect. And we're also going to see nation-states using these technologies because they're going to have the resources to not only do it well but to do it at scale. So there's going to be some very interesting types of attacks that are going to go on. And ultimately The result of that is our trust in digital communications, that's going to take a big step backwards, which means there may be issues for innovation and for investment and for adoption for different technologies that it's just going to put a pause in things a little bit, slow things down as part of our digital revolution. Maybe that's good. That's also, you know, potentially a downside. But deepfakes are entertaining at the moment. But make no mistake, when a cybercriminal can use it to impersonate someone's grandchildren and ask them for money, or forge the identity of a rival to bully them, to make them look foolish or to embarrass them, or perhaps to log in to a sensitive work meeting as one of the employees. Well, you kind of get the picture here, right? A lot of bad things can happen. The limits to how deepfakes could be maliciously used is really only limited to the creativity and the greed of the attackers, something that they have an overabundance of. We need to keep an eye on this innovative technology so that we don't become victims of it.
