Tengrinews.kz - Kazakhstan's specialists have developed a method to detect multiple types of fake photos and videos, presented at the recent Kolesa Conf-2024 IT conference. This innovation is entirely a domestic effort.
With deepfake detection becoming increasingly important, especially in finance, biometric verification - requiring users to show their face on camera - is a common way to secure transactions. However, criminals are now able to fake images to gain unauthorized access, creating a new challenge for advanced services. Kaspi.kz has developed its own technology to tackle this.
Photo: Kaspi.kz
In Kazakhstan, over 14 million people regularly use Kaspi.kz for financial transactions, document management, travel planning, and shopping.
"Before a user can access a product, service, or document, we need to confirm it’s genuinely them. Since the beginning of the year, there have been over 160 million biometric verification sessions in the Kaspi.kz app to protect our clients' data," explained Kaspi's Chief Data Officer, Duman Uvatayev.
Strict controls can make services inconvenient, so Kaspi opted to create their own security system instead of using existing solutions.
"To counter deepfakes, a combination of AI models is essential. We employ a range of neural networks at Kaspi, with specific architectures depending on data type, complexity of the fake, and detection accuracy. Our standards for accuracy are very high," Uvatayev said.
Kaspi.kz uses several types of neural networks — convolutional, recurrent, and others — to address specific detection tasks. Together, these models detect various types of deepfakes, ensuring that even if one network is deceived, another will catch it, allowing for robust security without compromising user convenience.
"This solution is a Kazakh product, created by local specialists who graduated from national universities. We’re constantly researching new technologies, and haven’t encountered such a complex combination elsewhere. We’re proud of our team's capability to create such unique IT solutions," Uvatayev added.
Currently, fraudsters often use different types of deepfakes: some swap one person’s face with another, others generate videos from photos, or sync lip movements to an overlaid voice.
Kaspi.kz's technology can detect all of these variations, ensuring data security for millions of Kazakhstani users while also enabling modern, user-friendly services.
AI-generated travel blogger has stirred up social media
