Research

Exploring AI Security, Trustworthy LLMs, and AGI Safety

Research Interests

LLM Reasoning Security

Detecting and mitigating backdoor attacks targeting the reasoning capabilities of Large Language Models.

Adversarial Machine Learning

Developing robust detection and defense mechanisms against adversarial attacks on vision and language models.

Trustworthy AI

Building reliable and safe AI systems for deployment in high-stakes environments.

AI Safety & Alignment

Researching systematic approaches to ensure AI systems behave as intended and remain aligned with human values.

Publications

4 papers
1
Jan 2026

STAR: Detecting Inference-time Backdoors in LLM Reasoning via State-Transition Amplification Ratio.

Seong-Gyu Park, Sohee Park, Jisu Lee, Hyunsik Na, Daeseon Choi.

arXiv:2601.08511 (Submitted to ACL 2026)

View Paper
2
Jun 2026 (To be presented)

ASTRA: Adversarial Stealthy Trigger Reasoning Attacks for Black-Box LLMs

Seong-Gyu Park, Sohee Park, Daeseon Choi

The 30th Pacific-Asia Conference on Knowledge Discovery and Data Mining(PAKDD 2026)

3
Aug 2025

Adversarial Image Detection for Vision Transformers via Attention Map

Seong-Gyu Park, Hyunsik Na, Daeseon Choi

IEEE International Conference on Advanced Visual and Signal-Based Systems (AVSS 2025)

View Paper
4
Jan 2025

Motion based user authentification for metaverse.

Seong-Gyu Park, Gwonsang Ryu

Journal of the Korea Institute of Information Security & Cryptology (2024): 493-503.

View Paper