The Role of AI in OT and Why Secure Integration is More Important Than Speed

[Erkan Teskancan]

## The Place of AI in OT and the Importance of Secure Integration

Manufacturers are exploring artificial intelligence (AI) to improve operational decision-making, optimize processes, and predict equipment failures. In theory, this technology is expected to increase efficiency, reduce unplanned downtime, and provide tighter control over product quality. However, joint guidance from the National Security Agency (NSA) and the Cybersecurity and Infrastructure Security Agency (CISA) emphasizes that AI integrations must be carried out with extreme caution.

### The Importance of AI in the OT Environment

As CISA's acting executive director Madhu Gottumukkala states, "OT systems are the backbone of our nation's critical infrastructure, and integrating AI into these environments requires a careful, risk-informed approach." The use of AI should be aimed at strengthening essential services, not creating an open door for systems to be harmed. Therefore, governance, assurance, and risk management processes must be completed before AI is deployed in OT environments.

While poor analytical practices can lead to financial losses or planning problems in an office environment, incorrect AI outputs in operational environments can halt production, create legal risks, or endanger human lives.

### Risks and Challenges of AI in OT

AI introduces new forms of failure. Problems with data integrity quickly undermine the reliability of AI systems, models can drift, and decision-making mechanisms may not be very transparent. Furthermore, hackers can create malicious inputs to deceive or poison AI models. This poses a direct risk in a controlled environment, especially if decisions related to production control, quality thresholds, or safety are affected.

In OT environments, equipment lifespans are long, upgrade windows are limited, and downtime costs are high. While the digitalization process continues, many legacy systems are still in use, and their compatibility with modern platforms can be challenging. AI systems developed in isolated environments may behave differently when connected to live production data.

### The Place of AI within the OT Architecture

The position of AI within OT determines operational risk. Some systems offer recommendations, and a human operator makes the decision, while others can directly change process parameters or make autonomous decisions. These two modes of use require different approaches in terms of security and governance.

A consulting system involves the operator in the process and provides oversight; autonomous systems, on the other hand, create a direct dependency between the model and production. For example, an AI system recommending a maintenance schedule is relatively low risk, while a system that directly changes production parameters requires strict control and security measures.

### Assurance of Model Performance

AI models can lose performance in environmental conditions different from the data conditions used for training. Production environments are constantly changing: equipment ages, maintenance affects performance, product mix changes, and supply chains evolve. Without monitoring, models can deviate from real conditions but still appear reliable.

Therefore, manufacturers must establish structured processes to ensure performance monitoring, identification of model retraining needs, comparison of prediction accuracy with actual results, and regular auditing of model behaviors.

### Building a Practical Foundation

AI provides measurable benefits in production environments when managed carefully. Predictive maintenance, quality analysis, and process optimization benefit from this technology. However, all benefits depend on how the technology is integrated into operational systems.

Organizations must clearly understand AI's place in the OT architecture, its interaction with existing control systems, and cyber risk management. Implementation must be consistent with security engineering practices, cybersecurity measures, and operational governance. In AI implementation, priority should be given to disciplined integration rather than speed.

AI can support operational performance, but only when it operates within a framework that preserves safety, reliability, and system integrity. The NSA and CISA guidance emphasizes that once AI enters the production environment, it too becomes operational technology.