This paper analyzes 18 commercially available HAZOP software tools, evaluating their level of automation, cost/time efficiency, input requirements, output features, and AI integration — including LLM capabilities and prompt engineering. It examines how tools like SALUS HAZOP AI use LLMs and prompt engineering to automate information retrieval and hazard identification from internal databases of past incidents. Key challenges identified include hallucination, reduced accuracy when deviating from trained scenarios, and high costs. The authors propose a conceptual AI-assisted HAZOP framework integrating P&ID data with prompt-engineered LLM queries. Relevant to PSM because HAZOP is a cornerstone of Process Hazard Analysis under OSHA PSM, and this work directly addresses how prompt engineering can improve or undermine AI-assisted hazard identification.
