kaz29Вестник Казахстанско-Британского технического университетаHerald of the Kazakh-British Technical University1998-66882959-8109Казахстанско-Британский Технический Университет10.55452/1998-6688-2026-23-3-176-187kaz29-3183Research ArticleКОМПЬЮТЕРНЫЕ НАУКИCOMPUTER SCIENCEПРИМЕНЕНИЕ ГЕНЕРАТИВНЫХ МОДЕЛЕЙ ИИ ДЛЯ АТАК И ЗАЩИТЫAPPLICATION OF GENERATIVE AI MODELS FOR ATTACKS AND DEFENSEhttps://orcid.org/0009-0001-6017-1659БатырхановаА. А.BatyrkhanovaA. A.

М.т.н., сениор-лектор

Алматы

M.E.Sc., Senior Lecturer

Almaty

a.batyrkhanova@iitu.edu.kzhttps://orcid.org/0009-0004-0319-0636БекмуханА. С.BekmukhanA. S.

М.т.н., сениор-лектор

Алматы

M.E.Sc., Senior Lecturer

Almaty

a.bekmukhan@iitu.edu.kzhttps://orcid.org/0009-0005-2983-3377АбильдаеваT. T.AbildayevaT. T.

М.т.н., сениор-лектор

Алматы

M.E.Sc., Senior Lecturer

Almaty

t.abildayeva@iitu.edu.kzhttps://orcid.org/0000-0003-4270-1908ЕскендироваД. M.YeskendirovaD. M.

К.т.н., ассоциированный профессор

Алматы

Cand. Sc. (Tech.), Associate Professor

Almaty

d.yeskendirova@iitu.edu.kzМеждународный университет информационных технологийКазахстанInternational Information Technologies UniversityKazakhstan202625092026233176187Copyright © Батырханова А.А., Бекмухан А.С., Абильдаева T.T., Ескендирова Д.M., 20262026Батырханова А.А., Бекмухан А.С., Абильдаева T.T., Ескендирова Д.M.Batyrkhanova A.A., Bekmukhan A.S., Abildayeva T.T., Yeskendirova D.M.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://vestnik.kbtu.edu.kz/jour/article/view/3183

В статье рассмотрено двойное применение генеративного искусственного интеллекта в кибербезо­пасности: как инструмент атаки и как элемент защиты. Показано, что генерация правдоподобного текста, изображений и аудио упрощает злоупотребления (фишинг, дипфейки, подмена коммуникаций), а в задачах машинного обучения возникает отдельный риск малые, но целенаправленные возмущения входа. Прак­тическая часть выполнена на воспроизводимом стенде с классификатором рукописных цифр MNIST: на чистых данных модель демонстрировала стабильное распознавание, после чего применялась атака FGSM с варьированием параметра е. Зафиксировано, что визуально почти незаметный шум способен переводить уверенную классификацию в ошибочную (на характерном примере «7» была получена неверная метка «3»). Для противодействия проверены два реконструктивных подхода: обычный автокодировщик (AE) и деноизинговый автокодировщик (DAE), обученный восстанавливать чистый сигнал из зашумленного входа. При росте е DAE снижал долю ошибок заметнее, чем AE, поскольку лучше отделял высокочастотные компонен­ты возмущения от полезных штрихов. Дополнительно введен детектор аномалии по энергии реконструкции E(x), позволяющий не только «очищать» вход, но и фиксировать факт подозрительного воздействия для последующего анализа инцидентов.

This paper examines the dual role of generative artificial intelligence in cybersecurity: as an enabler of attacks and as a practical defense component. It is shown that realistic generation of text, images, and audio lowers the barrier for abuse (phishing, deepfakes, impersonation), while machine learning systems face a distinct risk class­small but targeted input perturbations. The practical part is implemented as a reproducible benchmark around an MNIST handwritten-digit classifier. After verifying stable performance on clean inputs, adversarial examples were generated using FGSM with multiple values of e. The results confirm that a visually subtle perturbation can flip a confident prediction to an incorrect class (in a representative case, a “7” was forced to be classified as “3”). Two reconstruction-based defenses were then compared: a standard autoencoder (AE) and a denoising autoencoder (DAE) trained to recover clean signals from noisy inputs. As e increases, DAE reduces the misclassification rate more noticeably than AE, which is consistent with its training objective of separating high-frequency noise from meaningful strokes. To make the defense actionable in an operational setting, a simple anomaly detector based on reconstruction energy E(x) was also introduced. This adds a second layer: the DAE attempts to restore the input for correct classification, while E(x) provides an explicit alert signal suitable for logging and incident review.

generative AIcybersecurityadversarial examplesFGSMMNISTautoencoderdenoising autoencoderanomaly detectionreconstruction errorgenerative AIcybersecurityadversarial examplesFGSMMNISTautoencoderdenoising autoencoderanomaly detectionreconstruction errorReferencesGoodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. Advances in Neural Information Processing Systems, 27, 2672–2680 (2014).Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. Advances in Neural Information Processing Systems, 27, 2672–2680 (2014).Hu, W., and Tan, Y. Generating adversarial malware examples for black-box attacks based on GAN. arXiv preprint (2017). arXiv:1702.05983.Hu, W., and Tan, Y. Generating adversarial malware examples for black-box attacks based on GAN. arXiv preprint (2017). arXiv:1702.05983.Xiao, C., Li, B., Zhu, J.-Y., He, W., Liu, M., and Song, D. Generating adversarial examples with adversarial networks. Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI) (2018), pp. 3905–3911.Xiao, C., Li, B., Zhu, J.-Y., He, W., Liu, M., and Song, D. Generating adversarial examples with adversarial networks. Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI) (2018), pp. 3905–3911.Lin, Z., Shi, Y., and Xue, Z. IDSGAN: Generative adversarial networks for attack generation against intrusion detection. PAKDD 2022, LNCS 13282 (2022), pp. 79–91.Lin, Z., Shi, Y., and Xue, Z. IDSGAN: Generative adversarial networks for attack generation against intrusion detection. PAKDD 2022, LNCS 13282 (2022), pp. 79–91.Lakshmanan, R. WormGPT: New AI tool allows cybercriminals to launch sophisticated cyber attacks. The Hacker News (15 July 2023).Lakshmanan, R. WormGPT: New AI tool allows cybercriminals to launch sophisticated cyber attacks. The Hacker News (15 July 2023).Haidar, A. Kazakhstan pushes nationwide AI rollout amid cybersecurity risks and skills shortages. The Times of Central Asia (18 August 2025).Haidar, A. Kazakhstan pushes nationwide AI rollout amid cybersecurity risks and skills shortages. The Times of Central Asia (18 August 2025).Afane, K., Wei, W., Mao, Y., Farooq, J., and Chen, J. Next-generation phishing: How LLM agents empower cyber attackers. arXiv preprint (2024). arXiv:2411.13874.Afane, K., Wei, W., Mao, Y., Farooq, J., and Chen, J. Next-generation phishing: How LLM agents empower cyber attackers. arXiv preprint (2024). arXiv:2411.13874.Roy, S.S., Thota, P., Naragam, K.V., and Nilizadeh, S. From chatbots to phishbots? Phishing scam generation in commercial large language models. Proceedings of IEEE Symposium on Security and Privacy (2024), p. 221.Roy, S.S., Thota, P., Naragam, K.V., and Nilizadeh, S. From chatbots to phishbots? Phishing scam generation in commercial large language models. Proceedings of IEEE Symposium on Security and Privacy (2024), p. 221.Croitoru, A., Ionescu, R.T., Khan, F.S., Shah, M., et al. Deepfake media generation and detection in the generative AI era: A survey and outlook. arXiv preprint (2024). arXiv:2411.19537.Croitoru, A., Ionescu, R.T., Khan, F.S., Shah, M., et al. Deepfake media generation and detection in the generative AI era: A survey and outlook. arXiv preprint (2024). arXiv:2411.19537.Saeed, U., Jan, S.U., Ahmad, J., Shah, S.A., Alshehri, M.S., Ghadi, Y.Y., Pitropakis, N., and Buchanan, W.J. Generative adversarial networks-enabled anomaly detection systems: A survey. Expert Systems with Applications, 296, 128978 (2026). https://doi.org/10.1016/j.eswa.2025.128978Saeed, U., Jan, S.U., Ahmad, J., Shah, S.A., Alshehri, M.S., Ghadi, Y.Y., Pitropakis, N., and Buchanan, W.J. Generative adversarial networks-enabled anomaly detection systems: A survey. Expert Systems with Applications, 296, 128978 (2026). https://doi.org/10.1016/j.eswa.2025.128978Samangouei, P., Kabkab, M., and Chellappa, R. Defense-GAN: Protecting classifiers against adversarial attacks using generative models. Proceedings of ICLR (2018).Samangouei, P., Kabkab, M., and Chellappa, R. Defense-GAN: Protecting classifiers against adversarial attacks using generative models. Proceedings of ICLR (2018).Heiding, F., Schneier, B., Vishwanath, A., Bernstein, J., and Park, P.S. Devising and detecting phishing emails using large language models. IEEE Access, 12, 42131–42146 (2024). https://doi.org/10.1109/ACCESS.2024.3375882Heiding, F., Schneier, B., Vishwanath, A., Bernstein, J., and Park, P.S. Devising and detecting phishing emails using large language models. IEEE Access, 12, 42131–42146 (2024). https://doi.org/10.1109/ACCESS.2024.3375882Heiding, F., Lermen, S., Kao, A., Schneier, B., and Vishwanath, A. Evaluating large language models’ capability to launch fully automated spear phishing campaigns: Validated on human subjects. arXiv preprint (2024). arXiv:2412.00586.Heiding, F., Lermen, S., Kao, A., Schneier, B., and Vishwanath, A. Evaluating large language models’ capability to launch fully automated spear phishing campaigns: Validated on human subjects. arXiv preprint (2024). arXiv:2412.00586.SlashNext. WormGPT: The new generative AI tool used by cybercriminals to launch business email compromise attacks (2023).SlashNext. WormGPT: The new generative AI tool used by cybercriminals to launch business email compromise attacks (2023).Europol Innovation Lab. ChatGPT: The impact on law enforcement (2023).Europol Innovation Lab. ChatGPT: The impact on law enforcement (2023).Rössler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., and Nießner, M. FaceForensics++: Learning to detect manipulated facial images. Proceedings of IEEE/CVF International Conference on Computer Vision (ICCV) (2019).Rössler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., and Nießner, M. FaceForensics++: Learning to detect manipulated facial images. Proceedings of IEEE/CVF International Conference on Computer Vision (ICCV) (2019).Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., and Fergus, R. Intriguing properties of neural networks. arXiv preprint (2013). arXiv:1312.6199.Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., and Fergus, R. Intriguing properties of neural networks. arXiv preprint (2013). arXiv:1312.6199.Goodfellow, I.J., Shlens, J., and Szegedy, C. Explaining and harnessing adversarial examples. arXiv preprint (2015). arXiv:1412.6572.Goodfellow, I.J., Shlens, J., and Szegedy, C. Explaining and harnessing adversarial examples. arXiv preprint (2015). arXiv:1412.6572.Vincent, P., Larochelle, H., Lajoie, I., Bengio, Y., and Manzagol, P.-A. Extracting and composing robust features with denoising autoencoders. Proceedings of the 25th International Conference on Machine Learning (ICML) (2008), pp. 1096–1103.Vincent, P., Larochelle, H., Lajoie, I., Bengio, Y., and Manzagol, P.-A. Extracting and composing robust features with denoising autoencoders. Proceedings of the 25th International Conference on Machine Learning (ICML) (2008), pp. 1096–1103.

The authors declare that there are no conflicts of interest present.