Paper Type
Complete
Paper Number
PACIS2026-1351
Description
Programmable financial infrastructures increasingly embed compliance logic into smart contracts and token standards, while regulation remains expressed in natural language. This study asks how auditability can be operationalised as a design objective in AI-augmented compliance systems for tokenised real-world assets. We present GRACE (Governance-aware Regulatory Auditability and Compliance Engine), a layered architecture integrating AI-augmented rule generation, validation, and human-in-the-loop governance. Using Design Science Research and Hong Kong AML regulation as the demonstration context, we derive five design principles and instantiate them through Knowledge, Authoring, and Governance layers. The evaluation provides directional evidence that structured expert intervention and heuristic reinforcement improve rule-level enforceability and verifiability while reducing structural intervention demand. The findings position governance-aware design, rather than model optimisation alone, as central to auditable programmable compliance.
Recommended Citation
Chung, Ming Hin; Bawm Win, Treza; and Zhong, Hao, "GRACE: An AI-Augmented Compliance Information System for Real-World Asset Tokenization" (2026). PACIS 2026 Proceedings. 3.
https://aisel.aisnet.org/pacis2026/blockfintech/blockfintech/3
GRACE: An AI-Augmented Compliance Information System for Real-World Asset Tokenization
Programmable financial infrastructures increasingly embed compliance logic into smart contracts and token standards, while regulation remains expressed in natural language. This study asks how auditability can be operationalised as a design objective in AI-augmented compliance systems for tokenised real-world assets. We present GRACE (Governance-aware Regulatory Auditability and Compliance Engine), a layered architecture integrating AI-augmented rule generation, validation, and human-in-the-loop governance. Using Design Science Research and Hong Kong AML regulation as the demonstration context, we derive five design principles and instantiate them through Knowledge, Authoring, and Governance layers. The evaluation provides directional evidence that structured expert intervention and heuristic reinforcement improve rule-level enforceability and verifiability while reducing structural intervention demand. The findings position governance-aware design, rather than model optimisation alone, as central to auditable programmable compliance.
Comments
06-Fintech