POLICY-ML: Policy-as-Code Enforcement for Compliance, Auditability, and Trust Across Data and Machine Learning Pipelines

Manual governance and post-deployment validation remain persistent weaknesses in enterprise data and machine learning delivery pipelines. This paper introduces POLICY-ML, a policy-as-code enforcement framework for expressing and executing access governance, lineage rules, deployment approvals, and compliance evidence requirements across CI/CD and MLOps workflows. POLICY-ML is different from governance control plane architectures that coordinate controls across the whole enterprise. Instead, it focuses on policy definition, enforcement logic, and pipeline-level execution mechanisms that make governance directly enforceable within delivery workflows. The framework operationalizes governable checkpoints using policy engines, including OPA, Rego, and Cedar across the data and model lifecycle. Operational experience in executable policy enforcement across enterprise AI delivery systems reduces compliance violations by 60–78%, improves audit evidence completeness by 39–49 percentage points, and cuts compliance reporting effort by 55–65% compared with manual or legacy validation workflows.