Digital Trust in Public Funds: An Explainable AI Framework for Continuous Compliance Monitoring of U.S. Federal Grant Subrecipients

Abstract

Background: Federal grants flow through thousands of state agencies and subrecipients, yet oversight remains largely periodic and document-driven. Two 2024 regulatory developments transformed this environment: the revised OMB Uniform Guidance (2 CFR Part 200, effective 1 October 2024) directs pass-through entities toward risk-based, continuous subrecipient monitoring and raised the Single Audit threshold to one million dollars, while OMB Memorandum M-24-10 (March 2024) established risk-management expectations for artificial intelligence in federal use — including requirements for explainability and human oversight. Together they create both the mandate and the guardrails for AI-assisted compliance monitoring. Objective: We design and specify an explainable AI (XAI) framework — the Grant Risk Assessment and Compliance Engine (GRACE) — for continuous monitoring of subrecipient compliance that couples cross-system anomaly detection with tamper-evident evidence preservation. Methods: Following a design-science methodology, we (i) analyzed the 2024 regulatory corpus (2 CFR 200, GAO fraud-risk guidance, NIST AI Risk Management Framework), (ii) derived a cross-system red-flag taxonomy spanning financial, payroll, procurement, and performance-reporting systems, (iii) specified a five-component architecture, and (iv) evaluated it through regulatory-requirement mapping and a fully synthetic case demonstration. Results: GRACE maps onto every major obligation of the 2024 Uniform Guidance monitoring cycle; the synthetic walkthrough shows how feature-level attributions (e.g., duplicate-invoice rate +1.82, payroll-to-performance mismatch +1.45) translate into a triaged composite score (89/100 → site review) with a plain-language explanation and an immutable ledger entry for every monitoring event. Tamper-evident logging via hash chains and periodic Merkle-root anchoring gives auditors and pass-through entities a shared, verifiable evidence base. Conclusions: Explainable, provenance-preserving anomaly detection can convert subrecipient oversight from episodic document review into continuous, defensible, and auditable practice — a foundational mechanism of digital trust in the stewardship of public funds. Empirical validation on operational data is the required next step and is specified as such.