Botnet Simulation and Observation Framework for AI-Driven Virtual Personas

Abstract

The rise of AI-driven content on social media has enabled the creation of sophisticated political botnets — networks of automated accounts that mimic human users to amplify narratives. This paper presents a framework for Virtual Identities for Botnet Exploitation and Simulation that simulates a network of virtual political botnets in a controlled environment. The proposed system, implemented in Python, deploys multiple bots in a controlled Telegram chat environment; each bot is endowed with a unique profile, including ideological stance and communication style, and operates autonomously via an LLM-based text-generation engine. In an exemplary study, small swarms of agents produced echo chambers, synthetic consensus, and cooperative exchanges, with occasional adversarial turns – demonstrating that the framework can generate and capture behaviors relevant to moderation and bot-detection research. Results demonstrate that coordinated AI bots can engage in dynamic conversations, with observed behaviors such as self-reinforcement among like-minded agents and adversarial exchanges between opposing ones. These results have both practical and theoretical relevance. Practically, the platform can be used to develop countermeasures and detection algorithms in cybersecurity. Theoretically, it provides insight into how synthetic social actors might influence information ecosystems. The work also highlights important ethical considerations, underscoring the need for responsible AI deployment in social contexts.