EAGER: Securing AI Research: Developing and Validating a Lifecycle-Based Typology of Threats Through Critical Incident Analysis and Participatory Engagement

Virginia Polytechnic Institute And State University, Blacksburg VA

Investigators

Abstract

Artificial intelligence (AI) is transforming industries, education, and society. As its influence grows, so do the risks associated with AI research. This project explores how to better safeguard AI research from growing security threats, including data breaches, stolen intellectual property, and foreign interference, that could undermine its integrity and impact. By developing a new framework that maps out potential threats across every stage of the AI research lifecycle, it aims to help institutions, educators, and policymakers understand and address vulnerabilities in AI research more effectively and holistically. Through collaboration with stakeholders across disciplines and sectors and the use of innovative research methods, this work will not only advance the field of Research on Research Security (RoRS) but also support safer, more resilient AI innovation for the public good. AI research is increasingly recognized as both a strategic capability and a domain of significant vulnerability. As global competition intensifies, AI research faces a growing array of security threats, including theft of proprietary algorithms, unauthorized access to sensitive training data, premature dissemination of high-risk findings, and undue foreign influence. Despite the urgency of these issues, the emerging field of Research on Research Security encounters some persistent challenges: (1) limited access to empirical research security data; (2) the absence of a robust, interdisciplinary community of practice; (3) a lack of well-developed theoretical frameworks; (4) the need for more diverse and qualitative methodologies; and (5) insufficient discipline-specific research, particularly in AI. To address these gaps, this EAGER project investigates a fundamental question at the intersection of AI and research security: How can security threats be systematically identified and described across the lifecycle of AI research? Drawing on the expertise of diverse stakeholders and disciplines within RoRS, and leveraging a combination of qualitative, computational, and participatory methods, this project aims to develop and validate a lifecycle-based typology of threats for AI research through participatory engagement. Grounded in firsthand empirical data, this typology will inform institutional practices, training, and policy development to strengthen the security and integrity of AI research. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.