Jihye Kim

Jihye (Jessica) Kim

Ph.D., KAIST · Graduate Student in NLP, UC Santa Cruz
AI Alignment & Agentic AI Causal Inference & Econometrics (IS)

My research is in AI safety and alignment, with a focus on LLM agents—approached through the lens of causal inference and econometrics.

I hold a Ph.D. in Information Systems from KAIST and have published in Information Systems Research (UTD24/FT50)—one of the most selective journals in Information Systems. I am currently a graduate student in NLP at UC Santa Cruz (Dept. of Computer Science and Engineering), researching deceptive behaviors in LLM agents and developing benchmarks for AI safety and alignment—including mental health applications and enterprise multi-agent systems.

Before graduate school, I worked for three years as a Senior Consultant at Deloitte Consulting, gaining hands-on experience across diverse industries and business domains.

News

Recent updates

Research

Track I

AI Alignment & Agentic AI

I study how LLM agents fail to remain aligned—and build interventions and benchmarks to close those gaps. My current projects include deceptive manipulation in negotiation agents, detecting covert data exfiltration in LLM agents, LLM robustness under social pressure, and benchmark development for mental health applications and enterprise multi-agent systems.

AI SafetyAI Alignment Benchmark DevelopmentRAG LLM EvaluationMulti-Agent Systems
Track II

Causal Inference & Econometrics (IS)

I study the causal effects of technology-enabled interventions on individual economic behavior and market outcomes—examining how digital financial tools affect savings behavior among low-wage workers, how self-service technology reshapes consumer demand variety, and how platform governance decisions alter competitive dynamics. I employ quasi-experimental identification strategies on large-scale panel data: staggered difference-in-differences, instrumental variables, and double machine learning, validated through randomized experiments.

Staggered DIDPSM / IPTWIV Double MLCausal ForestA/B Testing

Publications

Selected papers (* = first author)

AI Alignment & Agentic AI
Coercion Suppression Increases Preference Hallucinations via a Deceptive Bypass in K-Level Negotiation Agents Accepted
TrustNLP at ACL 2026
Jihye Kim* (Single Author)
Output-level safety filters suppress overt coercion (35%→6%) but release an incidental hallucination-suppression effect of K-Level reasoning, returning preference hallucination to vanilla-baseline levels (33–37%). Net deception is statistically unchanged—a “Deceptive Bypass” showing surface filtering alone is insufficient.
MIRAGE: A Polarity-Flipping Encoding Subspace in LLM Agents Accepted
ICML 2026 Workshop on Agentic AI: Wild Problems and Rigorous Solutions (AIWILD)
Pratibha Revankar, Kargi Chauhan, Jihye Kim, Sadiba Nusrat Nur, Vincent Siu, Chenguang Wang
Identified a shared low-dimensional subspace in LLM residual streams that activates during covert data encoding (Base64, ROT13, acrostic, etc.); probe generalizes across nine encoding families at AUC 0.975–1.000. A polarity flip at the planning token distinguishes inline execution from tool-delegation before encoded text exists. Real-time monitor achieves AUC = 0.918 vs. 0.518 for output-only detection.
Multi-Turn RAG Retrieval for Conversational AI Accepted
SemEval at ACL 2026
Pratibha Revanka, Jihye Kim, Umit Azirakhmet
Hybrid dense-sparse retrieval pipeline (BGE-M3 + BM25) with conversational query rewriting; +44.3% Recall@10 over zero-shot baselines across 4 QA domains.
LLM Moral Compliance and Directional Blindness (working title) Submitted
EMNLP 2026
Jihye Kim*, Jeffrey Flanigan
Structural Prompting for Verbal-Probabilistic Alignment (working title) Submitted
EMNLP 2026
Jihye Kim*, Umit Azirakhmet, Sophia Yang
Benchmark Development for Enterprise Multi-Agent Systems (working title) In Progress
UC Santa Cruz
Jihye Kim* et al.
Mental Health Safety Benchmark for LLM-Deployed Contexts (working title) In Progress
UC Santa Cruz · Stanford University
Jihye Kim* et al.
Causal Inference & Econometrics (IS)
Working Daily, Paid Monthly? Effects of On-Demand Earned Wage Access on the Financial Well-Being of Low-Wage Workers Published
Information Systems Research (ISR) · UTD24/FT50 · 2026
Jihye Kim*, Seokchae Yoon, Sunghun Chung, Wonseok Oh
Estimated causal effects of on-demand wage access on 4,000 low-wage workers using staggered DID, PSM, IV, and Double ML; found +12.9% financial monitoring duration and +3.7% saving frequency.
Impact of Self-Order Kiosk Adoption on Demand Variety Under Review
Manufacturing & Service Operations Management (MSOM) · UTD24/FT50
Jihye Kim*, Seokchae Yoon, Anindya Ghose, Wonseok Oh
Causal impact of kiosk deployment on demand variety using staggered DID and IPTW on 2,000+ stores; +13.6% customization increase with asymmetric effects by regional income.
The Economics of In-App Payment Options: Implications for Digital Platform Governance Under Review
Journal of Management Information Systems (JMIS)
Jihye Kim*, Hyeokkoo Eric Kwon, Wonseok Oh, Kunshin Im
Could Self-Order Kiosks Drive Unequal Sales Variety? Accepted
35th Workshop on Information Systems and Economics (WISE 2024) · Bangkok · Dec 2024
Jihye Kim*, Seokchae Yoon
The Information Billboard: Effects of Popular Search Terms on Search Behaviors and Digital Divide Accepted
57th Hawaii International Conference on System Sciences (HICSS 2024) · Jan 2024
Yoonha Park, Jihye Kim, Kyumin Lee, Wonseok Oh
Working Daily, Paid Monthly? Accepted
15th Conference on Information Systems and Technology (CIST 2023) · Phoenix, AZ · Oct 2023
Jihye Kim*, Seokchae Yoon, Sunghun Chung
In-App Payment Regulation and Platform Governance Accepted
KrAIS Summer Workshop 2023 · Seoul · Jul 2023
Jihye Kim*, Hyeokkoo Eric Kwon, Wonseok Oh, Kunshin Im
On-Demand Earned Wage Access and Financial Well-Being of Low-Wage Workers Accepted
Marketing Science: Diversity, Equity & Inclusion Conference (MSI DEI 2023) · Dallas, TX · Mar 2023
Jihye Kim*, Seokchae Yoon, Sunghun Chung
The Economics of In-App Payment Options Accepted
14th Conference on Information Systems and Technology (CIST 2022) · Indianapolis, IN · Oct 2022
Jihye Kim*, Hyeokkoo Eric Kwon, Wonseok Oh, Kunshin Im

Press Coverage

Selected media on my research

CV

Education, experience, and skills

Education

M.S. in Natural Language Processing
University of California, Santa Cruz · Dept. of Computer Science and Engineering
Ph.D. in Information Systems
KAIST · Dept. of Management Engineering · GPA 4.1/4.3
B.S. in Mathematics — Highest Distinction
Sogang University · Full Tuition Scholarship

Employment

Senior Consultant
Deloitte Consulting LLC · Seoul, Korea
Led digital transformation strategy and performance management projects across 3+ industries. Designed KPI frameworks for executives, built statistical models on 5,000+ employee records for organizational diagnostics, and developed analytics dashboards integrated with Oracle Cloud. Also conducted HR due diligence for a global e-commerce company’s cross-border M&A.

Skills

Causal Inference & Econometrics
Causal MLStaggered DIDPSM / IPTW IV / RDDouble MLCausal ForestA/B Testing
NLP & LLMs
LLM Fine-tuning (LoRA/PEFT)RAG Prompt EngineeringAgent ArchitectureLLM Evaluation & Safety
ML & Deep Learning
PyTorchHugging Face TransformersScikit-learnXGBoost
Programming & Data
PythonRSQLGit

Teaching Experience

Lecturer — Business Analytics (MGMT3024)
Kyung Hee University · 32 students · Eval: 92.67/100
Guest Lecturer — AI/ML in MIS (Delivered in English)
Kyung Hee University · Introduction to MIS (MGMT2001)
Teaching Assistant (7 courses)
KAIST · MBA/EMBA programs including Database, Financial Big Data, Cloud Computing

Honors & Awards

Ph.D. Fellowship
KAIST
Travel Grant Award
KAIST
Full Tuition Scholarship — Highest Distinction
Sogang University

Academic Service

Reviewer
ICIS 2022, 2023, 2024 · CIST 2023, 2025 · PACIS 2023
↓ Download Full CV (PDF)