Jiseok Bang

Mr. Jiseok Bang


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
5 October 2026 in Applied Sciences
Verified
Journal Article
Open access
PDF file
JBJiseok Bang
Jiseok Bang
Yohan LeeYohan Lee
Yohan Lee
SLSang Su Lee
Sang Su Lee
YCYongJe Choi
YongJe Choi
SLSeungkwang Lee
Seungkwang Lee
Dynamic ransomware analysis presents a dilemma: terminating a sample prevents later behavior from being observed, whereas allowing destructive functions to complete can corrupt the analysis environment and its evidence. Copy & Return addresses this problem by suppressing selected destructive effects at function boundaries while allowing the original process to continue. An ABI-compatible substitute is introduced via dynamic-library interposition or binary injection to redirect a dangerous destination, preserve a mutable buffer, or bypass a destructive transformation before returning control to the unchanged caller. To identify potential intervention targets, we implement a static-analysis pipeline that ranks candidate functions and presents the underlying evidence for target confirmation. We evaluate candidate rankings on 10 selected ELF ransomware binaries and conduct detailed runtime case studies of MBRLocker, Conti, and Cylance. For MBRLocker, the 512-byte disk payload is redirected to a dummy file without modifying the actual boot sector. For Conti and Cylance, the original contents of eight target files per sample are preserved while file renaming, metadata appending, and ransom-note creation remain observable, demonstrating evidence-preserving damage suppression. Separately, we use numerical features obtained through static analysis to train and evaluate a lightweight ransomware screening model based on histogram-based gradient boosting. On 1156 binaries comprising 126 ransomware and 1030 benign samples, the model achieves a mean F1-score of 0.9124 and a mean false-positive rate of 1.62% under nested grouped cross-validation; the serialized model occupies approximately 39.3 KiB. These results demonstrate complementary roles for function-level damage suppression, static target-identification support, and lightweight AI-based ransomware screening.