Minkyoo Song
Ph.D. Candidate in Electrical Engineering at the Network & System Security (NSS) Lab, KAIST
Daejeon, South Korea
Advisor: Prof. Seungwon Shin
I study the safety and security of large language models and data-driven security systems, with a focus on adversarial ML. My research identifies vulnerabilities in emerging AI paradigms and develops defenses for robust deployment, while applying AI to real-world cybersecurity problems.
📰 Publications [C]: conference, [J]: journal, [U]: under review
-
[C]
SafeMoE: Safe Fine-Tuning for MoE LLMs by Aligning Harmful Input Routing.
The Fourteenth International Conference on Learning Representations (ICLR 2026)
-
[C]
MoEvil: Poisoning Expert to Compromise the Safety of Mixture-of-Experts LLMs.
2025 Annual Computer Security Applications Conference (ACSAC 2025, Distinguished Paper)
-
[C]
Refusal Is Not an Option: Unlearning Safety Alignment of Large Language Models.
34th USENIX Security Symposium (USENIX Security 2025)
-
[C]
When LLMs Go Online: The Emerging Threat of Web-Enabled LLMs.
34th USENIX Security Symposium (USENIX Security 2025)
-
[C]
Claim-Guided Textual Backdoor Attack for Practical Applications.
The 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics (NACCL 2025 Findings)
-
[C]
Obliviate: Neutralizing Task-Agnostic Backdoors within the Parameter-Efficient Fine-Tuning Paradigm.
The 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics (NACCL 2025 Findings)
-
[C]
31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2025)
-
[C]
2024 IEEE Symposium on Security and Privacy (SP) (S&P 2024)
-
[J]
Elsevier Expert Systems with Applications (ESWA)
-
[J]
A Large-Scale Bitcoin Abuse Measurement and Clustering Analysis Utilizing Public Reports.
IEICE Transactions on Information and Systems
-
[U]
Invited to Major Revision at IEEE Transactions on Knowledge and Data Engineering (TKDE)
-
[U]
Submitted to 33rd ACM Conference on Computer and Communications Security (CCS 2026)
-
[U]
Submitted to 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2026)
-
[U]
To Make Each Account Count: Exploring Credential Data Breach Threats through Victim-driven Analysis.
Submitted to IEEE Transactions on Information Forensics and Security (TIFS)
🌐 Professional Experience
-
S2W · Research Intern, AI Team
- Developed an LLM-based content moderation framework for illicit drug jargon detection, capturing contextual and lexical cues.
-
Reviewer
- The Web Conference (WWW) 2024
- ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) 2024
- ACL Rolling Review (ARR) 2024, 2025
-
External Reviewer
- The Web Conference (WWW) 2025
- ACM Conference on Computer and Communications Security (CCS) 2025
- IEEE International Conference on Distributed Computing Systems (ICDCS) 2025
- International Symposium on Research in Attacks, Intrusions and Defenses (RAID) 2025
- Annual Computer Security Applications Conference (ACSAC) 2025
-
KAIST Data Intelligence Laboratory · Undergraduate Research Intern
- Conducted big data mining on COVID-19 datasets to surface actionable insights.
-
KAIST Data Mining Laboratory · Undergraduate Research Intern
- Investigated abnormal node detection in bipartite networks via butterfly counting.
🎓 Education
-
Korea Advanced Institute of Science and Technology (KAIST)
Ph.D. Student in Electrical Engineering, NSS Lab — Advisor: Prof. Seungwon Shin
-
Korea Advanced Institute of Science and Technology (KAIST)
M.S. in Electrical Engineering, NSS Lab — Advisor: Prof. Seungwon Shin
-
Korea Advanced Institute of Science and Technology (KAIST)
B.S. in Industrial & Systems Engineering; Double Major in Electrical Engineering
🏅 Honors & Awards
-
4th Prize, Cybersecurity Paper Competition
Poisoning Expert to Compromise the Safety of Mixture-of-Experts LLMs
-
2nd Prize, Cybersecurity Paper Competition
Graph-based Deep Learning Framework for Credential Stuffing Risk Prediction
-
4th Prize, Cybersecurity Paper Competition
Delexicalized Distant Supervision for Illicit Drug Jargon Detection
-
4th Prize, Cybersecurity Paper Competition
Understanding the Occurrence and Impact of Credential Data Breach
-
Cum Laude
-
Academic Achievement Award: Salutatorian
-
Dean’s List
🌐 Languages
-
Korean
-
English
✉️ Contact
For collaborations or inquiries, please use the channels below.