A novel approach for Anti-Cheating: AI-based behavioural analysis

The pervasive issue of cheating in online video games poses a significant threat to fair competition, player engagement, and the integrity of eSports. Traditional anti-cheating solutions, often reliant on static detection techniques and elevated privileges (e.g. kernel-level access), suffer from limitations in adaptability, intrusiveness, and resistance to evolving threats. This thesis introduces an innovative approach to cheat detection in online video games, based on player’s behavioural analysis for anomaly identification. Leveraging machine learning (ML) and Vision-Language Models (VLMs), the proposed system is designed to be game-agnostic and player-specific, capable of identifying anomalous behaviours in real-time ensuring applicability across different game categories. The proposed framework adopts a client-server architecture: a client-side module responsible for the continuous collection and streaming of multimodal data—including keyboard, mouse, and gamepad inputs, as well as audio and video streams—and a server-side authoritative module that constructs individual behavioural profiles and detects deviations through unsupervised learning models. Input data is structured as a multivariate time series and enriched with behavioural features derived from kinematic, spatial, and statistical analyses. The behavioural model is trained to learn the player’s normal activity and is incrementally updated with newly observed legitimate behaviour. When an anomaly is detected, a VLM interprets the corresponding visual and auditory gameplay context, generating natural language explanations to enhance the transparency and reliability of the decision-making process. The system is designed for real-time execution with a scalable, low-overhead architecture suitable for competitive scenarios such as eSports tournaments. Experimental results demonstrate the framework’s capacity to detect cheating behaviours and improve fairness and security in online gaming. This work highlights the feasibility of a proactive and adaptive anti-cheating detection strategy by combining behavioural modelling with multimodal Artificial Intelligence.