About

Hello, and thank you for visiting. I am an AI safety and governance researcher at the Responsible AI Department, Frontier AI Lab, KT Corporation, and Adjunct Professor in Business Analytics at the Graduate School of Technology and Innovation Management, UNIST. I received my PhD in Technology Management, Economics, and Policy from Seoul National University in 2024.

My research asks how AI systems fail quietly, and how those failures can be measured before they reach people. Three threads run through it: red-teaming and defence of retrieval-augmented generation systems, risk taxonomies and governance for high-risk and Physical AI systems, and value alignment and evaluation of large language model judges.

I work across the boundary between industrial deployment and academic evaluation. At KT I lead projects on adversarial security for agentic RAG services and on a structured risk taxonomy for Physical AI — now being applied to real urban settings with the MIT Senseable City Lab and the City of Seoul — while my academic work develops diagnostics that separate moral orientation from moral calibration in human and machine evaluators. Methodologically I draw on adversarial testing, alignment auditing, sentence-embedding analysis, patent analytics, and network science.

Current focus

• Physical AI risk assessment for urban deployment — a twelve-month project with the MIT Senseable City Lab and the City of Seoul, applying the Physical AI Risk Taxonomy to a live city and working towards a Korea-specific evaluation standard.
• Physical AI risk taxonomy — a structured evidence map of risks that arise when AI perceives, decides, and acts in the physical world.
• Agentic RAG security — black-box poisoning attacks that hijack multi-step reasoning chains, and the defences against them.
• Pluralistic value alignment — where LLM evaluators diverge from human interpretive communities, and why.
• AI knowledge ecosystems — mapping policy, science, and technology to locate governance gaps.

Background

Before my current role I studied AI innovation and national competitiveness, publishing in PLOS ONE, the Journal of the Knowledge Economy, and Technology Analysis & Strategic Management. That quantitative training — economic complexity, scientometrics, knowledge networks — now underpins how I measure risk and alignment at scale.

This work also travels into policy. In October 2026 I joined the KDI delegation to the OECD in Paris, meeting the STI AI and Emerging Digital Technologies Division and presenting on multi-dimensional AI risk classification at the KDI–OECD Expert Roundtable on AI, Innovation and Productivity. I speak within Korea's NRC national AI collaborative research programme, advise STEPI, KDI, and KOTRA, contribute to the GSMA Responsible AI Maturity Roadmap, and review for Scientific Reports, PLOS ONE, and others.

I enjoy building things with collaborators. If you are working on AI safety, evaluation, or governance and think there is a useful overlap, please write to me at deep1003 [at] snu.ac.kr.