AI-POWERED PHISHING DETECTION IN UZBEKISTAN: IMPLEMENTATION STRATEGIES AND CHALLENGES

Abstract

Phishing and related cyber threats are rising worldwide and have become a significant issue in Uzbekistan. The country's digital adoption is high (89% internet penetration as of 2024), and recent reports highlight surging cyberattacks and fraud (e.g., over 12 million attack attempts in 2024, with fraud complaints up 34%). AI-powered detection systems—using techniques like natural language processing (NLP), machine learning (ML), and fine-tuned language models—offer promising accuracy; for example, a recent fine-tuned language model achieved 97.5% accuracy on a phishing dataset, and classical ML methods have yielded comparable results. This study examines the potential implementation of such AI solutions in Uzbekistan's context. We review state-of-the-art anti-phishing algorithms and analyze local factors (language, internet usage, regulatory environment). Our analysis finds that while technically feasible—supported by Uzbekistan's national AI strategy and digital initiatives—effective deployment faces challenges: Uzbek is a low-resource language with limited corpora, and infrastructure or data limitations may impede training and deployment. We propose a framework combining multilingual NLP, continual learning, and integration with cybersecurity policies. The findings suggest AI-driven phishing defenses could substantially reduce risk, but success depends on building local datasets, computational capacity, and expertise. Future work should focus on creating Uzbek phishing corpora and pilot deployments to validate these approaches.