As artificial intelligence reshapes engineering practice, large language models (LLMs) are emerging as potential tools in the development of safety-critical systems. This paper investigates their role in the early stages of functional safety engineering, focusing on how they can support hazard analysis, safety goal formulation, and functional safety requirements from real-world system descriptions. Using structured case studies, LLM-generated safety artefacts are compared with those produced by experienced engineers and evaluated for relevance, complete-ness, and compliance with safety principles. The findings highlight both the capabilities and limitations of LLMs in this context. Rather than replacing human expertise, LLMs show promise as augmentative tools that enhance documentation, traceability, and knowledge transfer. This study contributes to understanding how AI can be integrated into safety workflows and lays the groundwork for future hu-man–AI collaboration in safety-critical engineering.