Crawling the web with the purpose of finding malicious sites is still an open problem given how the internet is a highly dynamic environment, this constant change makes it difficult to consistently traverse it, for this reason the application of artificial intelligence methods, in particular those belonging to reinforcement learning are studied in this thesis. A method inspired by advances in Machine Learning and classic reinforcement learning paradigms, such as multi-armed bandits, is presented and applied to an experimental environment. Such environment consists of a graph of the internet and was produced by the employment of a breadth-first search universal crawler that traverses the web and saves its travelled path, which corresponds to a snapshot of how sites are interconnected.