Abstract
The global anti-money laundering (AML) regime is failing. Trillions of dollars are laundered each year, yet governments detect only a fraction of that activity, even as financial institutions spend hundreds of billions on compliance. In the United States, AML regulation has evolved from a retrospective, prosecution-oriented framework into an expansive, preventative regime that measures inputs rather than enforcement outcomes. Although artificial intelligence has demonstrated potential to improve detection, firm-siloed AI systems introduce substantial financial, technological, and systemic risks, including prohibitive development costs, data- privacy constraints, and market concentration among third-party service providers. This Article argues that federated learning offers a more effective and resilient model for AI driven AML compliance. Federated learning allows financial institutions to collaboratively train a shared algorithm on locally stored and encrypted data without transferring sensitive information to a central repository. By preserving data heterogeneity while expanding the observable universe of illicit activity, federated AI can improve detection accuracy, reduce false positives, lower adoption costs, and mitigate systemic risk associated with model monocultures. The Article concludes that, despite unresolved governance and accountability challenges, federated AI presents the most promising path for aligning AML compliance with its core objective: preventing financial crime.
First Page
61
Recommended Citation
Matthias Connelly,
Can AI Fix Anti-Money Laundering? The Case for Federated Intelligence in Financial Crime Prevention,
4
Student J. Info. Priv. L.
61
(2026).
Available at:
https://digitalcommons.mainelaw.maine.edu/sjipl/vol4/iss1/5
