Artificial intelligence games are a very interesting tool for teaching Artificial Intelligence techniques. Competitors write programs for agents, which are supposed to complete a given task or fight against other agents. In order to achieve the best performance, programs may have to use advanced Artificial Intelligence methods. In this paper, we present a framework to build artificial intelligence games, using Abstract State Machines (ASM) for the specification of the rules of the games. Choosing ASM, we expect that the competitors will be able to understand clearly the semantics of the rules. The framework includes a compiler for an ASM-based language, allows complete control of the order of execution of agents and easy integration with graphical libraries.