Sumsub has unveiled an upgraded deepfake detection system designed to keep pace with AI-generated fraud by learning from new attack patterns in real time rather than waiting for periodic model refreshes. The company says the new approach is meant to address a fast-moving threat environment in which synthetic identities, manipulated documents and deepfake-driven verification attacks are becoming harder to spot with static tools.

According to Sumsub, the Adaptive Deepfake Detector analyses a wider set of signals than facial imagery alone, including documents, geolocation, IP addresses, device indicators, facial biometrics, liveness checks and behavioural patterns across multiple users. It is also built to flag presentation attacks, injection attempts, third-party involvement and poor-quality inputs such as motion blur, glare and unusual expressions.

The launch comes as fraud increasingly shifts towards layered, multi-step schemes. Sumsub says such attacks rose by 180% in 2025 and made up 28% of all fraud detected on its platform globally, a trend it links to the growing use of generative AI to create more convincing fake identities, voices and videos. In its wider fraud reporting, the company has also warned that synthetic identity document fraud has surged sharply in the US, with sectors such as fintech, healthtech and e-commerce especially exposed.

Nikita Marshalkin, Sumsub’s head of machine learning, said modern deepfakes can no longer be reliably identified by sight alone and argued that effective decision-making depends on analysing several signals at once, in real time. The company says its system can adapt to new fraud patterns without manual retraining, allowing it to respond within hours rather than the weeks or months often needed for traditional model updates.

Source Reference Map

Inspired by headline at: [1]

Sources by paragraph:

Source: Noah Wire Services