Russian IT Firm Develops Method to Remove Users’ Biometric Traces from Facial Recognition Systems
Russian technology developer Kryptonit, part of X Holding, has introduced a method that can selectively remove a person’s digital profile from facial recognition systems. The solution allows individuals to withdraw consent for the processing of their biometric data.
Biometric data includes unique human characteristics – from fingerprints to voice patterns. These identifiers are used to verify identity and typically rely on at least two parameters. According to the developers, modern facial recognition systems are designed so that a person’s digital facial representation remains embedded inside a trained model even after the associated data has been deleted from the system’s database.
Algorithms identify individuals through these embedded facial representations, which makes it technically difficult to comply with legal requirements requiring the termination of biometric data processing. In the event of a data breach, the risks increase significantly. Unlike passwords, biometric identifiers cannot be changed if compromised. Attackers could potentially use leaked biometric data to create synthetic identities or to bypass authentication systems that rely on facial recognition.
Dispersion Method Enables Machine Forgetting
Engineers at the Russian IT company addressed the problem by developing a new algorithmic approach. Instead of simply hiding biometric data, the algorithm modifies the internal logic of the recognition model so that it can no longer identify a specific individual. In effect, the system disperses the relevant biometric representation within the model. At the same time, the system’s ability to recognize other users remains unchanged.
According to the developers, tests of the dispersion technique on widely used public benchmark datasets – commonly employed to evaluate neural network performance – were successful. The system’s ability to detect the targeted biometric parameters dropped by up to 88 percent. At the same time, the overall accuracy of the facial recognition model remained stable. In the research paper “Machine Forgetting of Faces in Search Tasks Based on Embedding Dispersion,” authored by Kryptonit AI laboratory specialist Mikhail Zakharov, extensive experiments indicate that the proposed method provides stronger forgetting while preserving the model’s usefulness for search tasks.
Applications in Security and Access Control
The new technology could be deployed in video surveillance systems with facial recognition capabilities, access control systems and enterprise security platforms. One practical use case is removing biometric profiles of former employees from corporate security systems.
In the future, the technology may also be integrated into open industrial frameworks – open-source solutions widely used in computer vision systems. This could provide businesses and government organizations with a new tool for responsible biometric data management. In many existing systems, facial recognition algorithms can still identify individuals even after their photographs have been removed from a database because the biometric profile remains embedded within the trained model.
Strengthening Personal Data Protection
The new system developed by Kryptonit engineers represents a step forward in technologies designed to protect personal data and regulate biometric processing. For users, the technology creates a practical way to opt out of biometric identification, ensuring that recognition algorithms no longer detect their digital profiles.
The company’s solution could also become a foundation for similar tools developed by other organizations. It may encourage further research into privacy-preserving computer vision systems and support the broader development of responsible biometric technologies.
The technology may eventually be deployed in government infrastructure as well, including smart city surveillance systems, biometric authentication for public services and security systems used in banks, offices, industrial facilities and remote digital services.
The developers believe the technology also has export potential. Facial recognition systems are expanding worldwide, and the market for personal data protection technologies is growing rapidly. In many countries, legislation already recognizes the right to delete personal data. Tools that make it technically possible to exercise that right could therefore see strong international demand.
