Peer-reviewed review synthesizing 2015–2025 literature on how Machine Learning ("ML") is used for leak detection, HAZID, predictive maintenance, emergency shutdown support, and digital twins, whereby the ML can itself create safety-critical vulnerabilities in process industries. The authors propose a process-safety taxonomy of seven AI-induced risk categories—model uncertainty, data risk, model risk, operational risk, human risk, cyber-physical risk, and governance risk—and link representative failure modes to major accident hazards (fire, explosion, toxic release, and escalation). They argue that verification, validation, uncertainty quantification, adversarial resilience, safety-case evidence, and lifecycle governance must be treated together, including an “AI-SIL” concept. Highly pertinent to PSM because it frames OT/ML cybersecurity as a contributor to incident scenarios rather than as a standalone IT problem.
