Sumsub, a number one full-cycle verification platform that permits scalable compliance, launched Adaptive Deepfake Detector. The brand new mannequin tackles the prevailing situation of conventional offline options being unable to detect the most recent deepfake scams. In contrast to its predecessors, Sumsub’s deepfake detector successfully spots rising forms of refined fraud by means of its ML-driven detection instrument with prompt on-line self-learning upgrades.
Whereas the answer is launching globally, its relevance is particularly clear throughout Africa, the place fraudsters are shifting from low-effort scams to extra refined AI-enabled assaults.
This shift is mirrored in Sumsub’s Identification Fraud Report 2025–2026, which discovered that Tanzania recorded the best fraud price on the continent in 2025 at 5.0%, whereas Uganda recorded a fraud price of 4.7%. Côte d’Ivoire additionally noticed fraud rise by 51% year-on-year to 4.5%. In Kenya, regardless of an general decline in fraud, deepfakes already account for practically 10% of fraud makes an attempt, highlighting how AI-enabled fraud is changing into extra distinguished even in markets the place conventional fraud is being decreased.
In South Africa, this shift is already seen. The nation’s general fraud price declined by 31% year-on-year to 1.4% in 2025. Nonetheless, deepfake incidents elevated by greater than 269% YoY, exhibiting that AI-enabled impersonation is shortly rising as the subsequent frontier in South Africa’s digital id panorama.
Periodic mannequin updates reveal a systemic vulnerability, particularly that, between upgrades, which might take weeks or months to launch and implement, new threats can bypass defences and trigger actual harm to digital app customers and corporations. The important thing differentiator of Sumsub’s new instrument lies in its detection accuracy, which stems from steady mannequin studying from fraud alerts throughout a number of layers, permitting it to adapt inside hours, somewhat than weeks or months.
For companies working in digital finance, funds, crypto, iGaming and different high-risk on-line sectors, the findings spotlight a rising want for fraud prevention techniques that may adapt in actual time, somewhat than relying solely on periodic mannequin updates.
“In 2026, the menace panorama has advanced, demanding threat administration groups to reply with the next-generation fraud prevention fashions. Fashionable deepfakes can now not be detected by the human eye, and decision-making ought to be primarily based on a number of sign evaluation in actual time”, mentioned Nikita Marshalkin, Head of Machine Studying at Sumsub. “That’s why we launched our upgraded Deepfake Detector, providing shoppers not only a instrument, however somewhat a web-based studying system that mixes superior doc checks, system intelligence, and fraudulent networks evaluation to enhance deepfake detection capabilities. When the worth of failure is simply too excessive, a complete method to the growing AI-driven fraud problem is the reply we want”.
In present deepfake detection, threat groups can not rely solely on visible content material inspection. The complete context of the consumer session ought to be taken under consideration. Aside from producing deepfake pictures, voices or movies, fraudsters additionally use numerous injection strategies, thus offering a separate knowledge layer for prevention techniques to test and monitor.
From a technical standpoint, real-time detection primarily based on the ‘on-line studying’ mannequin implies no ready time for scheduled coaching cycles and no want for normal human overview to remain up-to-date.
As a substitute, the brand new resolution:
- Constantly learns new patterns, together with rising deepfake sorts or injection strategies, instantly incorporating them into the identified threats checklist;
- Indicators are collected from a number of sources, not on a single anomaly vector. The multilayered fraud detection system analyses paperwork, geolocation, IP handle, system alerts, facial biometrics (liveness) knowledge, and cross-checks verification data from a number of customers to identify fraudulent community exercise.
- Inside every new remark, the mannequin adjusts its parameters with no handbook retraining required.
- The detector’s resolution boundary shifts to account for evolving threats, pushing the typical detection accuracy near 100%.
To be taught extra about Sumsub’s Adaptive Deepfake Detector, please go to https://sumsub.com/deepfake-detection/
Picture credit score: Sumsub.
Supply: Sumsub.
