AI deepfake detection works by training artificial intelligence systems to recognize pixel-level patterns and unnatural inconsistencies that the human eye misses, such as irregular eye blinking or skin texture anomalies. However, this defense mechanism faces a critical challenge: each time a new deepfake generation method emerges, detection models require complete retraining, creating an endless cycle of technological catch-up .

How Do AI Systems Actually Detect Deepfakes?

The technology behind deepfake detection relies on understanding how artificial intelligence creates fake media in the first place. Generative Adversarial Networks, or GANs, work like a competition between two AI models: one tries to create fake images, and the other tries to detect them. As they "battle," both improve simultaneously. The faker improves at fooling, and the detector improves at catching fakes .

Researchers train separate AI models to recognize faces produced by well-known GAN systems such as StyleGAN2, StyleGAN3, ProGAN, and DCGAN, the same kinds of models behind many of the shockingly realistic fake faces seen online today. These detection systems analyze subtle inconsistencies that the human eye misses:

• Eye Blinking Patterns: Unnatural or irregular blinking sequences that don't match real human behavior
• Skin Texture Anomalies: Irregular skin texture under certain lighting conditions that reveal the generation process
• Pixel-Level Artifacts: Telltale digital fingerprints around the hairline and facial edges left behind by AI generation

These signals, invisible to most viewers, act as fingerprints left behind by the generation process itself .

Why Does Detection Become Obsolete So Quickly?

The fundamental problem is that detection models must constantly evolve. Back in 2022, researchers at Swinburne University of Technology Sarawak Campus trained classifiers to distinguish real faces from GAN-generated ones by learning these hidden patterns. Even then, the results showed how convincing these synthetic faces already were. Since then, the leap in generative AI has been staggering .

Tools that once required specialist computing power and weeks of training now produce photorealistic fake videos in minutes, accessible to almost anyone with a laptop. What was once a research novelty is now a tool misused for financial scams, misinformation, and identity fraud. Globally, fraud cases involving AI-generated voice and video impersonation have surged. In one widely reported case in the United Kingdom, a company lost over USD 240,000 after an employee was tricked by an AI-cloned voice of their CEO .

The challenge is that detection models must constantly evolve. Each time a new generation method emerges, detectors require retraining. It is an ongoing arms race that requires sustained research investment and collaboration between universities, industry, and policymakers .

How to Protect Yourself From Deepfake Scams

Technology alone does not solve this problem. Media literacy matters just as much. Here are practical steps everyone can take to avoid falling victim to deepfake fraud:

• Pause Before Sharing: If a video or image feels shocking or too dramatic, verify it through trusted news sources before forwarding it to others
• Look for Tell-Tale Signs: Unnatural blinking, mismatched lip movements, blurry edges around the face, or inconsistent lighting indicate a fake
• Use Verification Tools: Platforms like Microsoft's Video Authenticator and various open-source deepfake detectors are increasingly available for public use
• Be Skeptical of Urgency: Scam calls and videos often create panic to override better judgment, so slow down and verify before acting

The risk extends beyond financial loss. Deepfakes spread false narratives, damage reputations, and create non-consensual intimate imagery, causing real harm to real people. Closer to home, Malaysians have encountered AI-generated audio used in scam calls, a tactic increasingly reported alongside the notorious Macau scam variants that continue to plague the country .

"The deepfake challenge is not going away, but neither are the researchers working to counter it," explained Dr. Khaled Elkarazle, a researcher with the Faculty of Engineering, Computing and Science at Swinburne University of Technology Sarawak Campus.

Dr. Khaled Elkarazle, Researcher, Faculty of Engineering, Computing and Science, Swinburne University of Technology Sarawak Campus

At institutions like Swinburne Sarawak, students engage with these problems, building the technical foundations to contribute to a safer digital future. The next generation of AI researchers, cybersecurity experts, and responsible technologists may well be sitting in a classroom right now, developing the detection systems that will stay ahead of tomorrow's threats .

AI is neither inherently good nor bad; it reflects how we choose to use it. The same intelligence that creates convincing fakes also powers AI deepfake detection systems that expose them. The question is whether we invest in that work, support the research, and build the awareness required to stay one step ahead. So the next time someone sends you a shocking video, ask yourself: can you trust what you see?