The rise of fake news has been the subject of several research studies in the last decade. This is due to the increasing number of Internet users and the simplicity in posting news over platforms and websites. Hence, researchers have been developing machine learning (ML) models to detect fake contents and warn readers. However, there is a limited number of Arabic fake news datasets in terms of articles and news sources. This paper aims at introducing the first large Arabic fake news corpus which consists of 606912 articles collected from 134 Arabic online news sources. An Arabic fact-check platform is used to annotate news sources as credible, not-credible, and undecided. Moreover, different ML algorithms are used for the detection task. Experiments show that deep learning models perform better than traditional ML models. Models training showed underfitting and overfitting problems which indicate that the corpus is noisy and challenging.