Third Generation AI Deepfakes: What Magicians Have Known for Years & Why Looking For Visual Cues Is No Longer An Effective Strategy
- The White Hatter
- 22 minutes ago
- 5 min read
For years, here at The White Hatter, we have encouraged youth, teens, parents, and caregivers to be cautious when interacting with people they meet online. One of the most common pieces of advice we shared was simple, if someone refused to join a live video conversation, that hesitation could be an important warning sign that they were not who they claimed to be. Today, we believe that advice needs to change.
The rapid advancement of artificial intelligence has fundamentally altered how we verify identity online. The reality is that we have now entered what we would describe as the third generation of AI powered deepfake technology, and many of the techniques we previously recommended for identifying fake video interactions are no longer reliable (1). As difficult as it may be to accept, the ability of the average person to visually determine whether someone on a live video call is genuine is disappearing at an astonishing pace.
When the first generation of AI deepfakes emerged, there were often noticeable flaws. We taught young people to look for irregular blinking, facial distortions, poor lighting transitions, unnatural facial movements, or lips that did not quite synchronize with speech. These visual imperfections were often enough to raise suspicion that something was wrong.
As second generation deepfakes became more sophisticated, many of those obvious tells disappeared. Fortunately, there were still a few practical challenges that could help expose a fake. We often recommended asking the person to hold three fingers in front of their face, rapidly move their hand across the screen, or open their mouth and move their tongue from side to side. These actions frequently confused the AI models of the day because they required complex movements that were difficult to generate convincingly in real time. Unfortunately, those tests are now also rapidly becoming obsolete.
The latest generation of AI video models has become remarkably capable of reproducing complex facial expressions, tongue movements, finger positioning, head turns, changing lighting conditions, and other real-time interactions with a level of realism that would have seemed impossible only a short time ago. In many cases, there is no longer a practical visual challenge that an average person can perform during a live video conversation that will reliably distinguish a real person from an AI generated one.
Perhaps the biggest change is psychological rather than technological. For years, reluctance to appear on camera was often considered suspicious. Criminals frequently avoided live video because they could not convincingly fake their identity. Today, many of them are eager to turn their camera on. In fact, the live video call itself may now be part of the deception. The very thing that once increased trust can now be used to manufacture it.
Those creating AI deepfake technology for criminal purposes are continuously studying the techniques used to detect them. Every time researchers, journalists, cybersecurity professionals, or educators identify a new visual clue, developers train the next generation of AI models to eliminate that weakness. It has become an ongoing technological arms race, and the pace of improvement is accelerating.
During a recent discussion we had, digital identity expert Mark Kramer summarized this challenge exceptionally well:
“Every visual trick we invent becomes a training label for the next generation of models. We are running a race where the finish line moves the moment we cross it. Blinking. Lighting. Head turns. Three fingers. Each one worked, briefly, then became a solved problem for GenAI. The real question is not, Can we design a visual test that fools the model? It is, ‘Can we stop relying on human perception to make trust decisions in a world where GenAI has already outpaced it?”
Unfortunately, we believe Mark has accurately described where we now find ourselves. Human perception is no longer keeping pace with artificial intelligence. While researchers continue developing sophisticated AI detection systems and digital identity verification tools, the average parent, caregiver, educator, or teenager should not assume they can simply “spot the fake” with their own eyes.
This reminds us of something professional magicians have understood for generations. Most people have heard the saying, “The hand is quicker than the eye.” Interestingly, experienced magicians will tell you that this is not actually how great magic works. Successful illusionists are rarely moving their hands faster than people can see. Instead, they are directing, or misdirecting, the audience’s attention exactly where they want it to be, while the real method remains unnoticed elsewhere. Magic is not simply about hiding the truth, it’s about shaping what people believe to be true. That lesson has never been more relevant than it is today given AI deepfake technology.
Modern AI generated deception is no longer trying to fool your eyesight, it’s trying to influence your judgment. Criminals understand that if they can create enough confidence in what you think you are seeing and hearing, they can influence the decisions you make. The video itself is often only the delivery mechanism, however, the true objective is persuasion.
This represents an important shift in how we should educate young people. Rather than teaching them to search for visual imperfections when it comes to deepfakes, we believe we should increasingly be teaching them to examine the purpose behind the communication itself. Instead of asking, “Is this person real?” we should also be asking, “What decision is this person trying to get me to make?”
Are they asking for money?
Are they requesting intimate images?
Are they encouraging secrecy from parents or trusted adults?
Are they asking you to move the conversation to another platform?
Are they creating urgency that discourages you from slowing down and thinking?
These behavioural questions remain remarkably effective because they focus on manipulation rather than appearance.
This is why critical thinking has become one of the most important digital safety skills we can teach. Even if a video call is authentic, even if the person is genuine, and even if the technology itself is real, the decision you are being encouraged to make can still be manipulative or harmful. Decision integrity may soon become more important than identity verification alone.
As parents and caregivers, this means helping our kids understand that trust should never be based solely on what they see on a screen. Instead, trust should be built through context, established relationships, independent verification, thoughtful questioning, and taking time before making important decisions and posting anytime online no matter what the platform or privacy setting.
The uncomfortable reality is that the days of asking someone on a video call to perform a few simple actions to prove they are genuine are rapidly coming to an end. Those challenges can now be replicated, automated, or convincingly simulated by advanced AI systems.
That does not mean we should become fearful or distrust everyone we meet online. Rather, it means we must evolve our digital literacy alongside the technology itself. Just as we taught previous generations how to recognize phishing emails, misinformation, and online scams, we must now teach the next generation that seeing should no longer automatically mean believing.
In a world where artificial intelligence can increasingly imitate reality, we now believe that our greatest defence will not be sharper eyesight, it will be stronger critical thinking!
If you would like to further reinforce the message in this article, share it with your youth or teen and then watch the accompanying short video together. It can help turn the information into a meaningful conversation and provide another opportunity to discuss how these issues may apply in their own online experiences. (The real person is on the right and the AI generated video is the one on the left)
Digital Food For Thought
The White Hatter
Facts Not Fear, Facts Not Emotions, Enlighten Not Frighten, Know Tech Not No Tech
References:
