AI in OT Sparks Cascade of Complex Challenges

According to Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency, and the Australian Signals Directorate's Australian Cyber Security Centre, a joint government advisory sets out four key principles for OT owners and operators: to understand AI, assess AI use in OT, establish AI governance, and embed safety and security.

The piece notes that while the guidance is directionally correct, OT environments often lack the trust foundation needed to implement it, with concerns over nondeterministic models like large language models and AI agents clashing with OT’s demand for stability and safety.

Experts emphasise issues such as model drift, explainability gaps, and new attack surfaces, and warn that trusted data workflows—including firmware authentication and signed updates—are frequently weak in OT, increasing risks for AI-driven decisions. Attackers’ potential to exploit AI to mask malicious activity and to find zero-day vulnerabilities is highlighted, underscoring the need for multiple lines of defence across prevention, detection, and recovery.

Overall, the article argues AI in OT offers upside for anomaly detection but could deepen long‑lived architectural debt if hygiene and segmentation are not fixed first.