AI-driven Cyber Attacks and Detection: A Comprehensive
Review
Mithrananda M., Nemsara S., Gamage E., Bandara T., Vishwajith O., Batuwitaarachchi
J., Priyanath K., Ethpitiya T., Ravinath C. and Fernando S.
Department of Computer Security and Network Systems,
Faculty of Computing,
NSBM Green University,
Mahenwaththa, Pitipana, Homagama, Sri Lanka
fernandorawini@gmail.com
fernandowrs@students.nsbm.ac.lk
Abstract
Artificial Intelligence (AI) is increasingly reshaping the landscape of cybersecurity by
enabling systems to autonomously detect, predict, and respond to a wide array of digital
threats. This paper investigates the comparative effectiveness of various AI-powered
detection techniques including supervised machine learning, deep learning, anomaly
detection, heuristic-based models, and graph neural networks. Each technique is
evaluated based on its strengths, limitations, and resilience against adversarial tactics
such as data poisoning and evasion attacks. Publicly available datasets such as
CICIDS2017, DARPA TC 2000, and CADETS were used to benchmark detection
performance using standard metrics like accuracy, precision, recall, F1-score, and false
positive rate. The study also evaluates federated learning as a privacy-preserving
solution in decentralized environments, such as Internet of Things and mobile systems.
A key contribution of this paper is the inclusion of a tabulated comparison of these
technologies to support the final discussion and aid the reader in understanding
operational trade-offs. Additionally, this research addresses the ethical and legal
implications of deploying AI in cybersecurity. These include concerns over algorithmic
bias, transparency, data privacy, and accountability in autonomous decision-making.
While AI enhances threat detection capabilities, it also introduces new risks related to
misuse, surveillance, and unjustified profiling if not properly governed. Legal
frameworks such as the General Data Protection Regulation (GDPR) emphasize the need
for explainability and consent in automated systems, which many current AI-based tools
struggle to meet. The paper concludes by advocating for a multi-layered, ethically
aligned defence framework that combines technical performance with legal compliance
and societal trust. By synthesizing technical evaluation with ethical considerations, this
work contributes a balanced view of AI’s promise and peril in modern cybersecurity
operations.
Keywords: Artificial Intelligence, Cybersecurity, Intrusion Detection, Federated
Learning, Ethics
1
2
