Agentic AI in OT: New Capabilities, Emerging Risks, and Governance Challenges

How AI in OT Transforms Raw Data into Autonomous Action

In the fast-evolving landscape of modern manufacturing, industrial leaders are fundamentally rethinking their operational architectures, moving away from passive monitoring tools and embracing active system control. This evolution is driven by the realization that interconnected facility environments require more than just raw telemetry; they demand context and real-time coordination. Implementing industrial agentic AI allows operations to bridge the “context gap” that leaves raw data operationally useless, enabling systems to interpret conditions, apply complex rules, and execute corrective actions autonomously.

In the article “Agentic AI in OT: New Capabilities, Emerging Risks, and Governance Challenges,” published by the IEEE Computer Society and written by Roy Krans and Dan Idzikowski, the focus is on how industrial organizations must address the critical security, architectural, and AI governance challenges required to transition from passive dashboards to fully autonomous control layers. For facility leaders, adapting to these interconnected systems extends far beyond technical optimization. It requires closing operational context gaps while establishing strict boundaries around machine decision-making.

One of the primary hurdles in deploying industrial agentic AI is that traditional operational infrastructure often struggles to correlate fragmented data streams across different plant systems. To overcome these limitations, a modern facility must integrate AI into Operational Technology (OT) on open, interoperable architectures that utilize the digital thread. By unifying point-in-time metrics with continuous lifecycle data, AI in OT can surface hidden operational patterns, such as subtle ambient temperature shifts affecting product quality or electrical noise from welding equipment causing intermittent faults, that human operators or isolated dashboards simply miss.

As AI in OT pushes connectivity deeper into legacy equipment, it exposes new attack surfaces that amplify AI governance challenges. Managing these risks demands a multi-tiered security strategy by segmenting networks, mapping active protocols, and routing legacy hardware through hardened gateways to keep plant systems safe from external threats. Simultaneously, addressing AI governance challenges requires setting clear, risk-based boundaries on agent autonomy. Because mathematical models can break down at physical extremes, maintaining human oversight in high-risk environments remains essential regardless of how reliably industrial agentic AI performs within standard operating ranges.

Ultimately, the trajectory toward full autonomy relies on an incremental progression through a structured maturity model. Deploying industrial agentic AI begins with continuous monitoring and recommendation logging, using digital twins and virtual commissioning to build organizational trust before allowing machines to update control logic independently. Proactively establishing frameworks to handle these emerging AI governance challenges ensures that technological capabilities do not outpace oversight, empowering organizations to scale autonomous operations safely and effectively.

Read the full article here.