A Security-by-Design Framework for Bounded Agentic AI in ERP-Integrated Smart Semiconductor and Electronics Manufacturing

Semiconductor and electronics manufacturers increasingly operate enterprise resource planning (ERP) systems, manufacturing execution systems (MES), and Industrial Internet of Things (IIoT) infrastructure. These platforms operate as cyber-physical systems rather than as standalone business tools. Their expanded connectivity and automated operations increase the attack surfaces. Simultaneously, conventional ERP automation remains largely rule-based and parameter-driven, providing auditability but limiting adaptability. Recent advances in multiagent systems and agentic AI suggest a shift toward ERP systems that act as autonomous decision layers across IoT-connected semiconductor fabs and electronics manufacturing lines. However, introducing such autonomy fundamentally alters the security and trust assumptions of existing ERP and MES architectures. This paper proposes a fivephase security-by-design methodology and a reference architecture for migrating from rulebased ERP automation to bounded agentic AI. The methodology specifically targets secure manufacturing and supply chain assurance by bounding agentic ERP authority over production planning, supplier selection, and yield-critical workflows in IIoT-connected semiconductor fabrication plants (fabs) and electronics manufacturing service (EMS) lines.