They demonstrate that facial recognition systems identify real faces after modifications made with AI

The facial recognition systems They are able to identify a person even if the image of their face has been heavily modified with generative artificial intelligence (AI) tools, which suggests that this technology is not based only on superficial visual similarities.

Generative AI is capable of drastically modifying a person’s appearance until it becomes unrecognizable, with results so realistic that they would not raise the suspicions of whoever sees the image. This capacity is especially worrying in a context of cyberattack or disinformation.

The cybersecurity company Kaspersky has tested facial verification systems, to verify if they are able to detect the modificationseven the most pronounced ones. The result of this work has been exhibited at HORIZONS, its annual reference conference in Europe.

Kaspersky’s Global Research and Analysis Team (GReAT) used a open source facial analysis tool based on computer vision and AI which, as explained in a press release, is used in research on facial recognition and automated image analysis systems.

The experiment sought to verify whether this facial analysis tool was capable of correctly identifying the identity that was hidden behind the modified images, in ten cases that have covered the use of generative AI to aging and rejuvenating facesand even for modify them completely until offering a different visual delay.

The company claims that the facial recognition system has been able to identify the original identity in all cases. “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,” Kaspersky said.

Furthermore, the company emphasizes that “even when facial appearance changes significantly, recognition algorithms can continue to detect persistent biometric traits that remain stable after transformations synthetics”.

For the principal security researcher in Kaspersky’s Global Research & Analysis Team, Maher Yamout, the results of this experiment show that “AI-generated facial transformations can preserve biometric identity even when human perception interprets the images as completely different people,”