AI detection of drug activity in Telegram bots and groups for security applications

Abstract

Encrypted messaging apps have become a center for illegal activities, especially drug trafficking, because of their privacy features. Telegram, in particular, offers anonymous group chats, channels, and bots that share information about drug supplies around the world. This paper presents a system that uses AI to detect drugrelated content on Telegram. We collect data from publicly available Telegram groups and bot chats, prepare and standardize the information, and apply natural language processing (NLP) and machine learning to flag illegal drug activity. Our work includes a detailed analysis of message patterns. We focus on keywords related to substances and posting behaviors, along with examples of interactions between dealers and buyers. A key aspect of our approach is customizing modern NLP and ML methods for Telegram's environment. This allows users to form pseudonymous groups without revealing their phone numbers. We measure the rate of drug offers compared to requests, identify the most common coded keywords, and compare our method to existing monitoring on platforms like WhatsApp and Reddit. Finally, we tackle challenges such as coded language, temporary messages, and policy limitations while discussing ethical concerns. Our findings show that AI-driven content analysis can significantly help law enforcement monitor encrypted drug markets on Telegram.