Global trends and governance insights in AI safety: a bibliometric analysis

This study provides a comprehensive bibliometric analysis of AI safety through a Scopus-indexed literature analysis between 1995 and June 2025, with VOSviewer applied to generate bibliometric maps and network visualizations of co-authorship, keyword co-occurrence, and citation metrics. AI safety research experienced rapid growth starting from 2020, with 81.58% of all publications emerging over the last five years. The research field of AI safety primarily relies on computer science, engineering, and mathematics, with the United States and the United Kingdom serving as its primary contributors. Research clusters in the field encompass technical areas such as reinforcement learning and adversarial robustness, alongside ethical governance and long-term risk. The study reveals disciplinary and geographic gaps, underscoring the need for global participation in this field. Quantitative analysis shows 3,971 citations, resulting in an h-index of 32 and a g-index of 46. Collaboration analysis indicates an average of 3.55 authors per paper, with the strongest co-authorship networks linking the United States, the United Kingdom, and Germany. The research provides valuable insights into current trends and suggests additional areas for investigation. The research establishes a vital, data-driven framework that supports researchers, policymakers, and funding agencies in advancing AI safety studies while creating comprehensive, responsible AI deployment strategies. The study provides empirical evidence to inform future interdisciplinary studies and help establish AI governance approaches and promoting the development of responsible AI safety worldwide.