Abstract
To map the applications and challenges of the use of artificial intelligence ( AI ) in occupational health, a scoping review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews ( PRISMA-ScR ). Five databases—PubMed, Web of Science, EBSCOhost, IEEE/IET Electronic Library, and ACM Digital Library—were searched for English-language studies from January 1, 2015, to November 12, 2025, focusing on large language models ( LLMs ) and generative AI. Selection and extraction were made by a single reviewer, and there was no assessment of protocol registration or risk of bias. Results were narratively synthesized by technical features. Of the 131 records, 84 were screened after deduplication; 25 were included (18 articles, 7 conference proceedings ). The designs of the selected studies comprised experimental studies (n = 8), performance evaluations (n = 7), reviews (n = 5), conceptual papers (n = 3), and randomized controlled trials (RCTs; n = 2). LLM-based chatbots were reported as delivering individualized health information, detecting stress, and supporting decision-making, while computer vision ( CV ) systems were employed to detect personal protective equipment use and unsafe behaviors. In education and training, domain-specific LLMs improved motivation, summarization ability, and safety literacy. The evidence remains preliminary; key risks include hallucination, algorithmic bias, privacy concerns, and occlusion issues in CV. AI appears promising as an assistant for workers, managers, and occupational-health professionals; its implementation should be guided by organizational AI-governance regulations and professional ethics. Funding: none.
