Method of Counteracting Manipulative Queries to Large Language Models

Скорочена інформація про документ
dc.contributor.authorKovalchuk, Yehor
dc.contributor.authorKolomytsev, Mykhailo
dc.date.accessioned2026-03-03T10:34:52Z
dc.date.available2026-03-03T10:34:52Z
dc.date.issued2025
dc.description.abstractThe integration of Large Language Models (LLMs) into critical infrastructure (SIEM, SOAR) has introduced new attack vectors, specifically prompt injection and jailbreaking. Traditional defense mechanisms, such as input sanitization and Reinforcement Learning from Human Feedback (RLHF), often fail against semantic obfuscation and indirect injections due to their inability to distinguish between control instructions and data context. This paper proposes a novel method for detecting manipulative prompts based on a Multi-Head DistilBERT architecture. Unlike standard binary classifiers, the proposed model decomposes the detection task into four semantic vectors: malicious intent, instruction override, persona adoption, and high-risk action. To address the scarcity of labeled adversarial datasets, we implemented a hybrid data generation strategy using Knowledge Distillation, employing a superior model (Teacher) to label synthetic attacks for the compact Student model. Experimental results on both synthetic and real-world datasets demonstrate that the proposed system achieves a Recall of 0.99, significantly outperforming traditional TF-IDF and keyword-based baselines. The solution operates effectively as a middleware layer, ensuring real-time protection with low computational latency suitable for deployment on edge devices
dc.format.pagerangeP. 114-118
dc.identifier.citationKovalchuk, Y. Method of Counteracting Manipulative Queries to Large Language Models / Yehor Kovalchuk, Mykhailo Kolomytsev // Theoretical and Applied Cybersecurity: scientific journal. – 2025. – Vol. 7, No. 3. – P. 114-118. – Bibliogr.: 16 ref.
dc.identifier.doihttps://doi.org/10.20535/tacs.2664-29132025.3.345389
dc.identifier.urihttps://ela.kpi.ua/handle/123456789/79184
dc.language.isoen
dc.publisherIgor Sikorsky Kyiv Polytechnic Institute
dc.publisher.placeKyiv
dc.relation.ispartofTheoretical and Applied Cybersecurity: scientific journal, Vol. 7, No. 3
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/deed.uk
dc.subjectlarge language models
dc.subjectprompt injection
dc.subjectjailbreaking
dc.subjectnlp security
dc.subjectdistilbert
dc.subjectadversarial machine learning
dc.subject.udc004.8:004.056
dc.titleMethod of Counteracting Manipulative Queries to Large Language Models
dc.typeArticle

Файли

Контейнер файлів
Зараз показуємо 1 - 1 з 1
Ескіз
Назва:
114-118.pdf
Розмір:
12.63 MB
Формат:
Adobe Portable Document Format
Завантажити
Ліцензійна угода
Зараз показуємо 1 - 1 з 1
Ескіз недоступний
Назва:
license.txt
Розмір:
8.98 KB
Формат:
Item-specific license agreed upon to submission
Опис:
Завантажити

Зібрання

Theoretical and Applied Cybersecurity: scientific journal, Vol. 7, No. 3