Sumsub, a leading full-cycle verification platform focused on scalable compliance, has unveiled its Adaptive Deepfake Detector, a next-generation solution designed to combat increasingly sophisticated AI-driven fraud.
The launch addresses a major weakness in traditional offline detection systems, which often struggle to identify newly emerging deepfake scams. Powered by machine learning and real-time self-learning updates, the new detector can rapidly identify evolving fraud patterns with far greater accuracy.
Although the solution is being rolled out globally, its significance is particularly evident across Africa, where cybercriminals are moving beyond basic scams toward more advanced AI-enabled fraud tactics.
According to Sumsub’s Identity Fraud Report 2025–2026, Tanzania recorded the continent’s highest fraud rate in 2025 at 5.0%, followed by Uganda at 4.7%. Côte d’Ivoire experienced a 51% year-on-year increase, bringing its fraud rate to 4.5%. In Kenya, while overall fraud levels declined, deepfakes already make up nearly 10% of all fraud attempts, underscoring the growing role of AI-powered deception even in markets making progress against conventional fraud.
South Africa is also seeing this trend intensify. Although the country’s overall fraud rate dropped by 31% year-on-year to 1.4% in 2025, deepfake-related incidents surged by more than 269%, highlighting how AI-driven impersonation is rapidly becoming a major challenge within the country’s digital identity ecosystem.
Traditional fraud detection systems often rely on periodic model updates, leaving gaps that can persist for weeks or even months before new threats are addressed. During this window, emerging scams can evade detection and expose both businesses and users to significant risks. Sumsub says its Adaptive Deepfake Detector overcomes this challenge through continuous machine learning that analyses fraud signals across multiple layers, enabling the system to adapt to new threats within hours instead of weeks or months.
The findings highlight a growing challenge for businesses in digital finance, payments, crypto, iGaming, and other high-risk online sectors, where fraud prevention systems must increasingly respond to threats in real time rather than depend on periodic updates.
“In 2026, the threat landscape has fundamentally changed, forcing risk management teams to move toward next-generation fraud prevention models. Modern deepfakes are now virtually impossible to detect with the human eye, which means decision-making must rely on real-time analysis of multiple fraud signals,” said Nikita Marshalkin.
She added that Sumsub’s upgraded Deepfake Detector was designed not just as a detection tool, but as a continuously learning system that integrates document verification, device intelligence, and fraudulent network analysis to enhance deepfake prevention capabilities.
“When the consequences of failure are severe, tackling AI-driven fraud requires a comprehensive and adaptive approach,” Marshalkin said.
The company noted that effective deepfake detection now extends beyond analyzing visual content alone. Fraudsters increasingly combine AI-generated images, videos, and cloned voices with injection methods that create additional data layers for prevention systems to track and verify.
Technically, Sumsub’s “online learning” model enables real-time threat detection without waiting for scheduled training cycles or extensive human review.
The system continuously updates itself with newly identified fraud patterns, including emerging deepfake and injection techniques, incorporating them into its threat intelligence almost instantly. Rather than depending on a single anomaly signal, it analyses multiple data points simultaneously, including documents, geolocation, IP addresses, device signals, facial biometrics, and linked user verification activity to identify coordinated fraud networks.
As new observations are processed, the model automatically recalibrates its parameters without manual retraining, allowing its detection boundaries to evolve alongside increasingly sophisticated fraud attempts and pushing detection accuracy close to 100%.
Go to TECHTRENDSKE.co.ke for more tech and business news from the African continent and across the world.
Follow us on WhatsApp, Telegram, Twitter, and Facebook, or subscribe to our weekly newsletter to ensure you don’t miss out on any future updates. Send tips to editorial@techtrendsmedia.co.ke
