|
Original research AI-DRIVEN ANOMALY DETECTION FOR VALIDATED GXP CLOUD INFRASTRUCTURE MONITORINGPages 57-68 Abstract
Cloud platforms that host GxP-regulated operations increasingly use machine learning anomaly detectors to monitor validated infrastructure, yet these detectors are seldom brought under the change-control and qualification discipline applied to the systems they watch. This article argues that a detector whose output triggers or suppresses an alert leading to a change-control, incident or deviation record on a validated system is part of that system’s control loop, and that its lifecycle must therefore produce reviewable evidence. We separate what the regulatory sources require (a validated state, controlled configuration change, and audit trails for GxP-relevant records) from what this article proposes as practice (training-data provenance records, alert-to-audit-trail binding, and a promotion gate for detector versions). We then present a validation-aware architecture of four layers: evidence capture, a change-control gate, a human qualification checkpoint, and alert-to-audit-trail binding. The architecture is assessed with an explicit evaluation framework: eight criteria, each traced to a regulatory source or flagged as proposed practice, scored on a four-level maturity rubric, and supplemented by three audit-query traceability scenarios and a failure-mode analysis. The assessment is analytical, not empirical. It shows which criteria the design can satisfy by technical enforcement, which remain procedural, and where residual risk persists, most notably the tension between continuous retraining and controlled change. A protocol for the quantitative validation that this study does not provide is specified for future work.
Keywords:
GxP compliance, anomaly detection, cloud infrastructure monitoring, change control, audit trail, machine learning operations, computerised system validation, data integrity.
|