Minecraft tunnels for covert communications

Abstract

Although bots carry negative connotations in wider internet usage, video game bots can be helpful to users by automating rote tasks. In this work, we augment Minecraft bots to provide privacy by tunneling traffic to anonymity services. We then analyze our bot design using performance metrics and detection techniques, including the chi-square test on action distributions and Shannon’s Entropy on byte streams. These bots create game sessions and embed Internet traffic into Minecraft actions. Hidden Markov Models (HMMs) were trained on human gameplay. These HMMs were then embedded into the client and proxy to generate realistic action sequences. Using a Minecraft server to run the game enforces protocol rules and mitigates problems associated with mimicry-based circumvention. The prototype interfaces with several existing censorship circumvention tools, the HMMs generate Minecraft action sequences matching live gameplay sequences, and the prototype achieves a 350 kBit/s transfer rate. Other contributions include an action sequence dataset for Minecraft and examples of bot detection techniques.