Deploying machine learning models in sensitive fields like healthcare and finance now requires structured governance, prompting the introduction of the GRACE-AI framework.
Current operational machine learning frameworks excel at managing the technical lifecycle of AI models. However, they frequently fail to support compliance needs in strictly regulated environments.
To bridge this gap, researchers introduced GRACE-AI, which stands for Governance, Risk, Assurance, Compliance, and Explainability for Artificial Intelligence. The framework integrates operational machine learning with structured accountability mechanisms.
The architecture relies on four interconnected components designed to handle automated obligation extraction and risk management. The Regulatory Alignment Layer maps regulatory requirements directly into pipeline constraints.
Next, the Dynamic Risk-Tiering Engine assigns adaptive governance risk levels based on composite risk scores. The Continuous Assurance Pipeline handles drift detection, monitoring, automated recalibration, and anomaly tracking.
Finally, the Explainability-by-Design Module generates stakeholder-oriented and interpretable explanations for model outputs.
When evaluated using three benchmark datasets supporting four prediction tasks across healthcare and financial domains, the framework achieved an average accuracy of 96.38 percent.
The unweighted cross-task mean AUC reached 0.973, outperforming the strongest evaluated standalone baseline, a Transformer model at 93.60 percent, by 2.78 percentage points. It exceeded the mean of the five evaluated standalone baselines by 6.06 percentage points.
GRACE-AI covered 94.7 percent of the risk and reduced compliance overhead by 38.6 percent. Drift detection operated with an average latency of 1.8 seconds.
While the findings confirm the effectiveness of the governance model in a controlled experimental setting, validation under operational, real-world conditions with human involvement and at production scale remains the next objective.
