VTB Bank to Develop “Ethical Censor” AI Model with Moscow State University
The Russian bank is building a neural network designed to monitor other neural networks — ensuring that AI assistants in financial services stay factually accurate and ethically sound.
VTB Bank, in collaboration with researchers from Moscow State University, is developing an artificial intelligence model dubbed the “ethical censor.” The system will review the responses generated by other neural networks within the bank’s digital services to ensure they meet ethical standards and contain no factual errors. The project aims to enhance both the reliability and trustworthiness of AI-powered customer interactions in the financial sector.
According to Vadim Kulik, Deputy Chairman of VTB Bank’s Management Board, the institution already employs several large language models across its digital ecosystem. The bank’s voice assistant, Leo, handles up to 40 types of customer operations. Its digital avatar, Tamara, provides face-to-face consultations via video, while an AI-driven investment advisor has already attracted more than one million users.
Earlier, speaking at the Kazan Digital Week forum, Kulik noted that the bank uses around 1,500 AI models across all its business processes. Integrating neural networks has generated tens of billions of rubles (approximately $120–150 million) in net income. Looking ahead, he sees the bank’s future in open banking — a unified system that allows users to manage accounts across multiple financial institutions through a single app.
