The rapid growth of Artificial Intelligence (AI) applications necessitates the development of sophisticated risk management models to balance its immense opportunities with significant risks. An AI system is defined as a machine-based framework that learns from data to make predictions or decisions, capable of influencing both physical and virtual environments. This ability to actively shape its surroundings distinguishes AI from traditional software, presenting a double-edged sword: while it can drive innovation and efficiency, it can also lead to negative consequences like biased decision-making, privacy breaches, and unintentional harm. Consequently, AI risk management has emerged as a crucial framework for measuring, managing, and mitigating the potential harms that AI can cause to individuals, organizations, and the environment.
A comprehensive bibliometric analysis by Bernardelli and Giudici (2025) provides a systematic mapping of the state-of-the-art research in AI risk management, revealing its key themes, trends, and conceptual structure. The study aims to organize the existing knowledge and highlight the pressing need for a quantitative framework to manage AI risks effectively. The analysis defines AI risk broadly, encompassing a wide spectrum of potential negative outcomes, from operational failures and ethical biases to security threats, societal disruptions like job displacement, and long-term existential threats from advanced AI.
Mapping the Research Landscape
To capture the interdisciplinary nature of AI risk, the authors conducted a bibliometric review using the Scopus database, which is known for its comprehensive indexing of peer-reviewed literature across fields like computer science, law, ethics, and business. The search focused on documents from 2015 to 2025 containing the terms “AI risk” or “Artificial Intelligence risk” in their titles or keywords. The year 2015 was chosen as the starting point, as it marks the beginning of significant academic and regulatory engagement with AI governance. The final dataset consisted of 129 documents, including conference papers, journal articles, and books, from 354 authors across 103 different sources.
The analysis reveals a dramatic increase in academic interest in AI risk, with a significant surge in publications in 2023 and a peak in 2024, highlighting the topic’s growing relevance. The research is moderately internationalized, with the United States leading in publication volume, followed by countries like China, Italy, and Germany, which show a strong tendency toward international collaboration. The United States and the United Kingdom serve as central hubs in the global research network, but distinct collaborative clusters have also emerged elsewhere, underscoring the interconnected nature of the global AI risk research community. The most influential sources for this research include a mix of technical conference proceedings and interdisciplinary journals such as AI and Society and the Computer Law and Security Review, reflecting the cross-sectoral nature of the discourse.
Core Thematic Structure of AI Risk Research
Through keyword analysis, the study identifies several core thematic clusters that shape the field. At the center are foundational principles related to ethics and governance. Keywords such as AI ethics, trustworthy AI, responsible AI, and AI governance underscore a dominant concern with the normative dimensions of AI. This is closely linked to policy-oriented terms like AI regulation and the EU AI Act, indicating strong scholarly engagement with the legal and institutional frameworks needed for responsible AI development.
A second major theme involves risk identification and security. This includes immediate technical concerns like cybersecurity, robustness, and safety, as well as broader, long-term threats highlighted by keywords such as existential risk and artificial general intelligence (AGI). The literature shows a clear focus on understanding both immediate operational dangers and more speculative, high-impact consequences.
The third core area focuses on evaluation and mitigation strategies. Keywords like risk assessment and impact assessment point to a strong emphasis on developing structured methodologies to systematically evaluate and manage AI-related harms. This reflects a push toward creating practical, operational tools for organizations to handle AI risks. The analysis also notes the emergence of new technological concerns, with terms like generative AI and large language models (LLMs) appearing frequently, signaling a recent shift in research focus.
Evolving Trends and Strategic Gaps
A thematic map analysis, which plots research themes based on their centrality (importance) and density (development), provides a strategic overview of the field’s structure. The map reveals that themes like digital transformation and explainability are “motor themes”—well-developed and central to the overall discourse. In contrast, topics such as AI safety, machine learning, and existential risk are identified as “niche themes”—internally coherent but relatively isolated from the broader conversation.
Critically, the analysis places themes like AI education and AI fairness in the “emerging or declining” quadrant, characterized by low development and centrality. Given their prominence in policy debates, this finding suggests a significant gap between societal concerns and the current academic treatment of these topics. Foundational concepts like AI risk management, AI ethics, and AI regulation are classified as “basic and transversal themes.” While highly central and important for structuring the field, they are not yet internally well-developed, indicating a need for deeper theoretical integration and methodological refinement.
This thematic configuration highlights the fragmented nature of AI risk research. Despite shared concerns, the field lacks a unified management approach, with disciplinary boundaries hindering integration. There is a notable misalignment between the rapid pace of AI innovation and the slower development of empirical tools and standards for risk measurement.
The Path Forward: A Call for Quantitative and Integrated Frameworks
The key takeaway from the bibliometric analysis is the urgent need for a quantitative, measurable, and practical approach to AI risk management. While ethical and regulatory discussions are prevalent, the field must move beyond principles to develop actionable risk indicators. The authors suggest a framework grounded in four key principles commonly found in risk management: Security, Accuracy, Fairness, and Explainability (SAFE).
- Security involves ensuring AI systems are robust and protected against cyber threats, aligning with traditional operational risks.
- Accuracy requires that AI outputs are reliable and valid, addressing what is known as model risk.
- Fairness introduces a crucial social dimension, demanding data representativeness and non-discrimination, a type of risk often externalized onto users and society.
- Explainability focuses on making AI systems understandable and auditable by humans, which is essential for governance and oversight.
By developing tools to quantify failures along these axes—such as the Key AI Risk Indicators (KAIRIs) proposed in related work—organizations can evolve toward creating AI systems that are not only technically sound but also ethically responsible and human-centered.
In conclusion, the study by Bernardelli and Giudici (2025) illustrates that AI risk management is a highly relevant but still formative field. It is characterized by conceptual fragmentation and a disconnect between normative principles and practical measurement. The research underscores the necessity of interdisciplinary collaboration to harmonize frameworks, fill empirical gaps, and develop integrated, quantitative risk assessment models. By providing a structured map of the research landscape, the analysis offers a valuable roadmap for researchers, policymakers, and practitioners working to build a safe, equitable, and responsible AI ecosystem.
Reference
Bernardelli, A. E., & Giudici, P. (2025). AI risk management: A bibliometric analysis. Risks, 13(7), 131. https://doi.org/10.3390/risks13070131