This paper explores using Large Language Models (LLMs), specifically ChatGPT, to assist human analysts in hazard analysis for safety-critical systems through a process the authors call co-hazard analysis (CoHA). Using STPA methodology, the authors systematically evaluated LLM responses across three increasingly complex control system scenarios. Results suggest LLMs can usefully support hazard cause elicitation when guided by structured prompts. This is directly relevant to MOC in PSM because every proposed process change requires a hazard review to assess whether new hazards are introduced. LLMs could accelerate MOC hazard screenings, help identify overlooked deviation scenarios, and provide a more thorough review of change impacts, particularly for organizations with limited access to experienced PHA facilitators.
