Against the backdrop of accelerating digitalization of the global economy and the expansion of electronic commerce, dispute management is transforming into one of the key factors of business financial sustainability. The most problematic manifestation is friendly fraud: in 2025, it accounts for up to 61% of all transactions for which disputes are initiated. Under these conditions, traditional approaches to claims handling, based on manual processing and a set of heuristic rules, demonstrate limited effectiveness due to the scalability of attacks and the increasing complexity of behavioral scenarios of abuse. This paper describes the specific features of applying modern artificial intelligence algorithms to automate dispute resolution, including graph neural networks (GNN) and large language models (LLM) integrated into a RAG architecture. The empirical basis is formed through the analysis of academic publications and industry reports, supplemented by an in-depth examination of the Social Discovery Group case, which makes it possible to substantiate the practical viability of hybrid AI solutions. The results obtained indicate that the implementation of automation can reduce dispute processing time from several hours to seconds, increase the win-rate coefficient in the digital goods segment to 72%, and simultaneously significantly reduce operational costs. Thus, the study is embedded in the context of the development of FinTech risk-management methodology and forms an architectural approach to designing autonomous systems oriented toward revenue protection.