Governing Healthcare AI and ML: Enforcing Column- & Row-Level Security on Sensitive Data

Abstract

The present paper discusses the way column- and row-level security can regulate sensitive healthcare information in AI and ML pipelines, trying to balance granular protection of privacy with the high utility clinical analytics. The presence of multitiered security architectures, FHIR-integrated cell-level controls, and model-agnostic cohort discovery platforms. A focused literature review revealed that the concept of fine-grained access control is feasible in practice, and also identified persistent vulnerabilities like linkage attacks and governance fragmentation. Based on the secondary research design, the study synthesised peer-reviewed articles, technical reports and implementation case studies to provide a thematic synthesis of architectural patterns, performance implications and compliance results of implementing attribute-level and record-level policies in health data clouds of scale. The results showed that column-level security was able to isolate high-risk clinical features, row-level security operationalised least-privilege and jurisdiction-conscious access, and integrated policy enforcement across ML lifecycles reduced configuration drift, provided better auditability, and enhanced regulatory readiness, though with platform maturity and adversarial robustness limitations. The general finding of the analysis was that granular security controls are an essential, yet not exhaustive, basis of credible healthcare AI, and ought to be coupled with sophisticated cryptographic measures, federated design, and dynamically shifting and zero-trust governance models to encourage sustainable tradeoffs between innovation and safety and patient data sovereignty.