This paper presents a novel framework, "HAZOP-GPT," which acts as a generative AI copilot to assist human experts in performing Hazard and Operability (HAZOP) studies. The study is directly relevant to Process Safety Management (PSM) as it details the engineering of a specific "Prompting Module." This module utilizes a sequence of four interconnected prompts designed to systematically 1) generate safety hazards using guide words, 2) identify causes, 3) identify consequences, and 4) suggest mitigation strategies. By breaking the complex HAZOP process into chained prompts, the authors demonstrate how prompt engineering can reduce the hallucination rate and increase the relevance of AI outputs in safety-critical workflows. The paper provides a concrete example of how "Chain of Thought" prompting techniques are applied to the rigorous, structured requirements of Element 9g (Hazard Identification & Risk Analysis).
