OpenRegAI: A Multi-Agent Framework for Explainable Financial Fraud Detection and Automated Regulatory Auditing

The rapid expansion of open banking ecosystems and emerging regulatory frameworks such as the EU AI Act have created new challenges for financial institutions: fraud detection systems must not only achieve high predictive accuracy but also provide legally interpretable explanations. Traditional black-box machine learning models perform well in identifying fraudulent activities but fail to meet regulatory transparency requirements. This paper proposes OpenRegAI, a cloud-native multi-agent framework that integrates graphbased fraud detection, explainable artificial intelligence, and automated regulatory auditing. The architecture consists of three coordinated agents: FraudNet, a Graph Attention Network (GAT) that captures relational fraud patterns in transaction networks; XplainAI, an explainability module based on GNNExplainer that generates interpretable decision rationales; and RegBot, a RetrievalAugmented Generation (RAG) system powered by a large language model for compliance auditing against financial regulatory texts. Experimental evaluation demonstrates that OpenRegAI achieves a PR-AUC of 0.88 under 0.13% fraud prevalence, significantly improving operational decision support. The framework reduces analyst decision latency by 59% while generating structured compliance narratives aligned with regulatory requirements. The proposed system establishes a practical pathway for accountable and regulation-compliant AI deployment in open banking environments