Researchers build AI system for cross-border data compliance

A new decision support system tackles the growing challenge of cross-border data flows under increasingly stringent data governance regimes. Researchers Yuepeng Zhou, Dongchi Xing, and Li Xiong model a firm's compliance decisions as a finite-horizon Markov decision process (MDP), where regulatory rules are converted into a computable minimal compliance mapping. Compliance is treated as a hard constraint rather than a penalty, which more accurately reflects real-world constraints. The system uses masked deep reinforcement learning to solve the MDP, with counterfactual path advantages providing interpretable signals.

Experiments show the learned policies outperform baseline approaches, delivering interpretable and auditable decisions. Key findings include that local data processing concentrates in states where small lawful transfers are not worth the compliance cost, and the localization boundary shifts systematically as regulations tighten. Credential acquisition is front-loaded within the compliance year, and shallow decision trees can replicate the policy's decisions with high fidelity. The system also reveals an "absorb-then-adjust" pattern: expected rewards decline before observable behavior changes, suggesting that firms may already bear significant burden even if they haven't changed their data transfers yet. Importantly, the system is regulation-agnostic and can be transferred to other jurisdictions and rule-based compliance problems.