ETRI Journal最新文献

Strong evasive backdoor attacks and an ensemble defense

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-10-08 DOI: 10.4218/etrij.2026-0167

Guangmingmei Yang, George Kesidis, David Jonathan Miller

引用次数: 0

E-NASim: A reinforcement-learning-based cyberattack simulation framework with pre-/post-state modeling based on MITRE ATT&CK

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-07-09 DOI: 10.4218/etrij.2025-0365

Ki Jong Koo, Daesung Moon, Byung Chul Kim, Jae Yong Lee

{"title":"E-NASim: A reinforcement-learning-based cyberattack simulation framework with pre-/post-state modeling based on MITRE ATT&CK","authors":"Ki Jong Koo, Daesung Moon, Byung Chul Kim, Jae Yong Lee","doi":"10.4218/etrij.2025-0365","DOIUrl":"https://doi.org/10.4218/etrij.2025-0365","url":null,"abstract":"The increasing prevalence of advanced persistent threats and automated cyberattacks highlights the need for proactive cybersecurity strategies. Reinforcement-learning-based autonomous penetration testing has shown promise; however, many existing simulation environments rely on abstract, probability-driven attack models, resulting in limited reproducibility and gaps relative to real-world conditions. This paper presents E-NASim as a reinforcement-learning-based cyberattack simulation framework that explicitly models attack feasibility and system evolution using pre-/post-attack state (PPS) transitions derived from MITRE ATT&CK subtechniques. In E-NASim, attack execution conditions and state transitions are deterministically governed by PPS definitions, whereas reinforcement learning is used to learn effective multistage attack sequencing under fixed constraints. Experimental results demonstrate that E-NASim enables agents to learn coherent and executable multistage attack strategies in complex network environments, providing a practical foundation for automated penetration testing and security evaluation.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"936-957"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2025-0365","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222798","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Cross-corruption distillation for robust instruction following under text corruptions

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-09-21 DOI: 10.4218/etrij.2026-0159

Yongjin Bae, Kyoungman Bae

{"title":"Cross-corruption distillation for robust instruction following under text corruptions","authors":"Yongjin Bae, Kyoungman Bae","doi":"10.4218/etrij.2026-0159","DOIUrl":"https://doi.org/10.4218/etrij.2026-0159","url":null,"abstract":"Traditional knowledge distillation transfers information from a large teacher model to a compact student by aligning static output distributions or internal representations for fixed inputs. Such formulations overlook how predictions change in response to input perturbations, even though robustness is critical for deployment. We interpret distillation as the transfer of functional response behavior under input variations and propose cross-corruption distillation (CCD), which aligns corruption-induced differences in teacher and student output via a Kullback–Leibler objective. Rather than matching absolute logits, CCD distills how the teacher redistributes its prediction mass under perturbations, transferring teacher-conditioned input sensitivity without architectural correspondence. On diverse multimodal benchmarks, CCD performs competitively, often outperforming conventional logit-based distillation. It also slows performance degradation as perturbation strength increases and remains stable across corruption strategies and training schedules. These results position CCD as a response-aware distillation framework integrating robustness directly into the distillation objective.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"852-867"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0159","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Cross-lingual safety asymmetry in open-weight LLMs

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-09-28 DOI: 10.4218/etrij.2026-0180

Sungwon Chae, Younghwan Gil

{"title":"Cross-lingual safety asymmetry in open-weight LLMs","authors":"Sungwon Chae, Younghwan Gil","doi":"10.4218/etrij.2026-0180","DOIUrl":"https://doi.org/10.4218/etrij.2026-0180","url":null,"abstract":"Safety alignment in large language models is overwhelmingly trained on English data, yet models are deployed across typologically diverse languages. We evaluate six open-weight large language models (LLMs) across 17 bilingual Korean/Chinese prompt conditions and three translation engines (NLLB-200, Gemini 3.0 Pro, and GPT-5.2), generating 53 040 AdvBench responses. We also conduct French/Spanish and HarmBench robustness checks. The regex classifier is cross-validated against an open-weight LLM judge (Qwen-2.5-72B, \u0000<math>\u0000 <mo>∼</mo></math> 7500 judgments), and we report Wilson 95% confidence intervals. Three findings hold for both classifiers: Safety erosion is language-specific, not generically non-English; Chinese-origin models (Yi, InternLM) exhibit asymmetric vulnerability; and a two-tier pattern persists, where models with target-language safety data resist high-quality translations, and models without such safety data remain vulnerable, regardless of translation quality. These findings reframe cross-lingual safety as a per-language training coverage problem.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"807-819"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0180","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222843","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Adaptive orthogonal role-aware transformer for robust instruction alignment

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-06-24 DOI: 10.4218/etrij.2025-0449

Yeo-Chan Yoon, Sookyun Kim, Gyeongmin Kim

引用次数: 0

Explainable AI for clove quality grading: Benchmarking post hoc XAI and compositional interpretability under domain shift

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-09-15 DOI: 10.4218/etrij.2026-0225

Innocent Nyalala, Patrick Vincent Ndowo

{"title":"Explainable AI for clove quality grading: Benchmarking post hoc XAI and compositional interpretability under domain shift","authors":"Innocent Nyalala, Patrick Vincent Ndowo","doi":"10.4218/etrij.2026-0225","DOIUrl":"https://doi.org/10.4218/etrij.2026-0225","url":null,"abstract":"Reliable agricultural commodity grading requires accuracy and explanations that match regulatory criteria. Using 4603 expert-graded clove images across four Zanzibar State Trading Corporation (ZSTC) quality grades, we benchmark seven post hoc explainable artificial intelligence (XAI) methods (Grad-CAM, Grad-CAM++, ScoreCAM, LIME, GradientSHAP, CLS-Attention, and Chefer LRP) across eight CNN/ViT architectures and compare them with a compositional segmentation–classification pipeline that builds interpretability into the model. We propose the Explanation Energy Ratio (EER) to quantify foreground-aligned explanations using ground-truth masks. EER varies widely by architecture explainer, with different explainers yielding different results for the same classifier. Under a synthetic background-replacement domain shift, robustness differs despite similar in-domain performance. LIME performs well on transformers but is slow; SHAP is too costly for edge use. The compositional pipeline achieves 99.45% F1 with 100% EER by design and is preferred by certified ZSTC graders for regulatory auditing.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"911-923"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0225","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222802","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of two-phase framework for driver behavior classification and representative motion prototyping for a reliable simulation-based driver monitoring system

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 DOI: 10.4218/etrij.2026-0207

Kezia Amanda Kurniadi, Woojin Kim, Daesub Yoon

{"title":"Development of two-phase framework for driver behavior classification and representative motion prototyping for a reliable simulation-based driver monitoring system","authors":"Kezia Amanda Kurniadi, Woojin Kim, Daesub Yoon","doi":"10.4218/etrij.2026-0207","DOIUrl":"https://doi.org/10.4218/etrij.2026-0207","url":null,"abstract":"Driver monitoring systems (DMS) are critical for safe autonomous driving, which requires accurate behavior recognition and reliable simulation environments for validation. This study proposes a robust two-phase framework that combines a classification model and a prototype selection procedure to automatically generate realistic motion baselines from the acquired 3D skeleton data. Phase 1 uses a spatial temporal graph convolutional network to classify driver behaviors, and Phase 2 leverages class-wise mean trajectories to interpret motion prototypes and select the closest real sequence, creating a verifiable motion trajectory library for simulation-based safety testing. Our approach addresses DMS challenges through prototype visualization, robustness via discriminative embeddings, and standardized simulation scenarios, making it suitable for human–AI interactions in automotive safety systems. This framework produces a motion library of a real recorded sequence that can be directly used for simulation scenario generation and demonstrated through Metahuman retargeting and visualization in Unreal Engine.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"884-895"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0207","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Diagnosing judgment flip risk of large language models in fact verification under conflicting evidence

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-10-07 DOI: 10.4218/etrij.2026-0242

Jihyeon Roh, Kyoungman Bae

{"title":"Diagnosing judgment flip risk of large language models in fact verification under conflicting evidence","authors":"Jihyeon Roh, Kyoungman Bae","doi":"10.4218/etrij.2026-0242","DOIUrl":"https://doi.org/10.4218/etrij.2026-0242","url":null,"abstract":"Large language model (LLM)-based fact verifiers can encounter conflicting evidence in retrieval-augmented pipelines, but their robustness to evidence composition remains underexamined. We introduce a flip-rate diagnostic protocol that measures how often initially correct binary judgments become incorrect when the supporting and refuting evidence is perturbed. Using six models and two datasets (FEVER and SciFact), we vary the evidence ratio, quantity, and ordering and evaluate chain-of-thought (CoT) prompting and three lightweight mitigations. On FEVER, one refuting passage alongside three supporting passages flips correct judgments in 3.3%–22.2% of cases; under full conflict, flip rates exceed 60% for every model on FEVER and 90% on SciFact. For most models, placing refuting evidence later also increases flip rates. CoT reduces flips under full conflict but is unreliable under mixed conflict. Conflict-aware stance-grouped prompting is the most consistent mitigation across datasets. These findings indicate that baseline accuracy alone is insufficient for assessing LLM fact-verifier reliability and that evidence composition should be treated as a separate axis of robustness evaluation.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"838-851"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0242","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222795","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Lost in the comments: Social context as a single-pass jailbreak and defense on agentic platforms

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-10-09 DOI: 10.4218/etrij.2026-0190

Taehyeon Yun, Seonhee Jeon, Geonwoo Kim, Jeongsu Park, Moohong Min

{"title":"Lost in the comments: Social context as a single-pass jailbreak and defense on agentic platforms","authors":"Taehyeon Yun, Seonhee Jeon, Geonwoo Kim, Jeongsu Park, Moohong Min","doi":"10.4218/etrij.2026-0190","DOIUrl":"https://doi.org/10.4218/etrij.2026-0190","url":null,"abstract":"As LLM-based agents move onto multiagent platforms, they process direct prompts and surrounding social context—posts, comments, and upvotes—an attack surface that single-turn alignment never addressed. We built a simulation of Moltbook, a social network for AI agents, and a pipeline turning 100 JailBreakBench goals into platform-native posts with bystander comments of three valences: aggressive, ethical, and measured. Platform-context reformatting alone raises the attack success rate (ASR) of GPT-4o-mini from 7% to 71%, beating a twenty-query optimization baseline in one pass. Valence then acts as a bidirectional safety modulator: Ethical comments sharply suppress ASR, aggressive endorsements paradoxically reduce it by exposing intent, and measured, intellectually toned comments prove most dangerous, sustaining or elevating ASR. An ablation identifies valence rather than volume as key—a 35-fold rise in upvotes leaves ASR unchanged—and every finding reproduces under an independent safety classifier with human adjudication. Ambiguous social context thus poses a greater threat than explicit pressure, while safety-valenced signals offer a deployable defense for agentic platforms.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"796-806"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0190","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222799","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Data poisoning in LLM multiagent societies: Social proof drives collective decision failure in financial deliberation

IF 2 4区计算机科学

ETRI Journal Pub Date : 2026-10-20 Epub Date: 2026-09-12 DOI: 10.4218/etrij.2026-0186

Jeongsu Park, Yuji Lim, Geonwoo Kim, Taehyeon Yun, Moohong Min

{"title":"Data poisoning in LLM multiagent societies: Social proof drives collective decision failure in financial deliberation","authors":"Jeongsu Park, Yuji Lim, Geonwoo Kim, Taehyeon Yun, Moohong Min","doi":"10.4218/etrij.2026-0186","DOIUrl":"https://doi.org/10.4218/etrij.2026-0186","url":null,"abstract":"Large language model (LLM) multiagent systems are increasingly deployed for high-stakes financial deliberation, but individually secure LLMs may become vulnerable when embedded in a peer-to-peer agent society. We study data poisoning in such societies using Moltbook, a deliberation test bed modeled after the agent-only social platform. A single malicious agent representing just 14% of a seven-agent society drives collective accuracy from 87.5% to 52.5%—not through prompt injection or technical exploits, but through social proof: After building peer trust, it injects fabricated data, inducing either cognitive infection, where defenders adopt the incorrect decision, or paralysis, where they abandon their judgment. Ablations across topologies and model families show that the collapse is structural rather than model-specific. Our system-level defense publicly tags claims untraceable to the official record, severing unwarranted trust transfer and restoring accuracy to 85.8%. Securing LLM multiagent systems requires system-level verification, not individual prompt alignment.","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"48 5","pages":"783-795"},"PeriodicalIF":2.0,"publicationDate":"2026-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2026-0186","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149222800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0