Kaspersky held HORIZONS, the company’s annual reference conference, where it showed that facial recognition systems can continue to identify a person even after Generative Artificial Intelligence (GenAI) tools drastically modify their appearance through facial aging or rejuvenation techniques. In some cases, the images generated looked like completely different people to the human eye, highlighting the robustness of modern facial recognition.
As facial recognition technologies become increasingly integrated into sectors such as security, border control, healthcare, finance and marketing, advances in generative AI are also facilitating the creation of highly realistic synthetic images and increasingly sophisticated facial modifications. Today, AI-powered applications are widely used to enhance images, retouch photographs, edit faces, and alter people’s appearance, posing new challenges for facial recognition systems when evaluating real identities against digital modifications.
To better understand how facial verification and facial recognition systems respond to these transformations, Kaspersky’s Global Research and Analysis Team (GReAT) conducted an experiment using an open source facial analysis tool based on computer vision and Artificial Intelligence, used in research on facial recognition and automated image analysis systems.
During the experiment, Kaspersky analysts processed original portraits using generative AI tools to simulate aging and rejuvenation scenarios. In many cases, the resulting images appeared to be of completely different people to human observers. However, despite these significant visual changes, the facial recognition system managed to correctly identify the modified images as belonging to the original identities in the ten cases analyzed, demonstrating the accuracy of facial recognition even in extreme conditions.
Advances in generative AI are also making it easier to create highly realistic synthetic images and facial modifications.
The experiment included AI-generated aging and rejuvenation scenarios, comparisons between visually very different portraits, and verifications using modern facial recognition software.
Geometric and structural characteristics
The results suggest that current facial recognition systems are based on deeper geometric and structural characteristics of the face, and not only on superficial visual similarities perceived by people. Even when facial appearance changes significantly, facial recognition algorithms can still detect persistent biometric features that remain stable after synthetic transformations.
From a cybersecurity perspective, the results highlight a double risk scenario. They demonstrate the ability of facial recognition-based authentication systems to resist certain forms of AI-driven visual manipulation. On the other hand, they raise important questions about the possible misuse of generative AI for identity theft, the creation of highly realistic fake profiles and the evasion of human verification processes, even in environments where facial recognition is the main security barrier.
“While the experiment does not represent a large-scale study, it does constitute a proof of concept for a potential AI-enabled cyberattack that the industry should reflect on. It demonstrates a critical practical implication: AI-generated facial transformations can preserve biometric identity even when human perception interprets images as completely different people. This creates new challenges for digital trust, facial recognition-based identity verification, and fraud prevention in an era marked by the rapid evolution of AI-generated synthetic content,” Maher explains. Yamout, principal security researcher in Kaspersky’s Global Research & Analysis Team.
As AI synthetic content generation technologies evolve, Kaspersky researchers emphasize that these advances require greater attention from digital identity system developers, cybersecurity professionals, and regulatory agencies to ensure that biometric technologies, including facial recognition, remain secure, reliable, and resilient in the face of new AI-driven threats.
