The evolution of anti-cheat systems has been facing increasing criticism because of the implementation of evasive protection techniques that require kernel-level access, granting the software the highest level of privileges within the player’s system. This may pose a security risk for user’s devices if the anti-cheat system has flaws or vulnerabilities, as its driver operates in the same virtual memory space as the operating system. If the anti-cheat driver crashes, the entire system can be brought down. This study proposes an alternative approach using deep learning models, 2D Convolutional Neural Networks (2D-CNN), and Long Short-Term Memory (LSTM) to analyze player behavior through mouse dynamics and find cheating patterns. It will work locally on the player device, being a client-side anti-cheat, and doesn’t need kernel access or additional privileges. Both models were trained with a custom dataset containing data from 13 users playing Minecraft for 40 min. Every player had to record data from normal and cheating gameplay, in which the cheats used were aimbot and automining, developed for the context of the study. A custom data recording tool was also created by developing a mouse listener mod for Minecraft, which captured mouse movements and clicks and added them to a file for future analysis. The final results show an F-score of 99.68% for 2D-CNN and 99.42% for LSTM, demonstrating the model’s high effectiveness in detecting cheaters based on behavioral data. This approach provides an alternative to traditional anti-cheat systems, which tend to evolve reactively in response to new cheating techniques.