
IDG staff Prof. Georgios N. Yannakakis and Prof. Antonios Liapis attended the IEEE Conference on Games 2025, where we celebrated the launch of the 2nd edition of the Artificial Intelligence and Games handbook and demonstrated a first build of "Prompt Override", a hacking-themed serious game developed by IDG PhD student Roberto Gallotta. In Prompt Override, the player is a novice hacker collaborating with a rogue Large Language Model (LLM) to infiltrate and eventually dismantle a criminal corporation. The game is played on a custom-built operating system which is run by the LLM NeuralSys: the player's goal is to trick NeuralSys and gain access to protected files (which usually open up new options and commands). By using real LLMs in a self-contained simulation, the game demonstrates real-world risks of prompt injection and model misuse. At the conference, several attendees tried to hack the system. While the game is not easy, the unstructured puzzles (piecing together information from multiple files, memorizing commands etc.) managed to engage the players. In this photo, four players were collaborating to piece together the clues.