TaxFL: Federated Learning and Federated Graph Intelligence for Cross-Border Tax Compliance

Cross-border tax non-compliance increasingly exploits fragmented data silos, complex ownership structures, and fast-moving digital assets such as cryptocurrencies. Traditional exchange-of-information (EOI) mechanisms, while essential for targeted cases, are poorly suited to the high-volume, pattern-based risk analytics required by modern compliance programs. This working paper proposes TaxFL, a pragmatic and privacy-preserving framework that enables tax authorities to collaboratively train and evaluate risk models without sharing raw taxpayer data. TaxFL integrates cross-silo federated learning, federated graph neural networks (GNNs) for ownership and transaction networks, and layered privacy-enhancing technologies (secure aggregation, differential privacy, and optional confidential computing). Beyond the technical stack, TaxFL provides a conservative "belt-and-suspenders" governance and legal blueprint, viable even under strict interpretations of international data-transfer rules (e.g., Schrems II). Preliminary validation on benchmark graph data (Cora) demonstrating robust federated convergence (>90% accuracy) despite extreme non-IID and privacy constraints, albeit with expected utility trade-offs compared to isolated local training. We outline four priority use cases, a realistic six-month pilot design for a small multi-jurisdiction consortium (starting with public/synthetic data), and a comprehensive evaluation framework covering detection performance, privacy assurances, operational overhead, and institutional trust. TaxFL offers a scalable, low-risk complement to existing EOI processes, providing a publishable blueprint for real-world deployment under appropriate safeguards. Open implementation available at https://github.com/pafrantz/TaxFL/tree/main. Seeking academic collaborators for rigorous empirical validation on real-world tax data under appropriate safeguards.Open implementation available at https://github.com/pafrantz/TaxFL/tree/main. Seeking academic collaborators for rigorous empirical validation on real-world tax data under appropriate safeguards.