How AI Should Be Used Safely in Heavy and Nuclear Industry

Summary

This opinion article argues that the industrial use of AI must be considered beyond software development, particularly as heavy industries such as automotive, energy and food production adopt AI-based decision agents and technical supervisors. The author is concerned that organizations may place language models in command dashboards, operational decisions or the creation of safety plans without fully understanding their behavior. In the author’s view, allowing AI to generate safety policies and then make decisions under those policies could create a feedback loop that moves AI into increasingly critical system components while weakening human understanding and established defense-in-depth practices. The article gives the nuclear industry special attention because nuclear power is entering a period of renewed development and because failures in this sector can affect surrounding systems and communities. It also frames vulnerable AI systems as a possible attack vector, citing jailbreaking and prompt injection as ways hostile actors might attempt to compromise industrial environments. The author rejects using fear, regulatory pressure or high barriers to entry as the main governance response. Instead, the article calls for open, clearly documented and auditable safety rules, collaboration across industry, and explicit limits on AI’s role. For future small modular reactors, Generation IV systems and existing plants, the author argues that AI should be integrated into established defense-in-depth policies rather than replacing them. The author discloses that AI tools translated, edited and formatted the English version, while the thesis and nuclear analysis remain the author’s own.