Deploying multiple AI systems across IoT and ICS/SCADA environments presents unique security challenges, as traditional IT-centric safeguards often don’t align with industrial operational realities. Organizations are advised to begin by anchoring their strategy in established frameworks—such as the “MAESTRO” model—for accountability, traceability and threat modelling, before selecting a primary AI platform and building developer sandboxes to safely test integrations. Ensuring that key AI components are run in controlled environments rather than on personal or unmanaged devices helps reduce risk, particularly in systems where industrial assets must remain continuously operational.

Further practical steps include enforcing robust access controls, limiting where AI agents and model-context protocol servers run; instituting mechanisms to securely ingest, validate and execute AI workflows; monitoring AI interactions across OT/IoT zones for anomalous behavior; and embedding standardized practices into supply chain and vendor workflows. By combining these measures with segmentation, secure sandboxing, and consistent governance across AI-enabled systems in industrial networks, organizations can better manage the risk of model hijacking, data leakage or operational disruption while pursuing innovation and automation in critical infrastructure.

‍

Source: https://gca.isa.org/blog/7-practical-steps-to-secure-multi-ai-deployments-for-iot-and-ics-scada