The detection of new Android malware is far from being a relaxing job. Indeed, each day new Android malware appear in the market and it remains difficult to quickly identify them. Unfortunately users still pay the lack of real efficient tools able to detect zero day malware that have no known signature. The difficulty is that most of the existing approaches rely on static analysis coupled with the ability of malware to hide their malicious code. Thus, we believe that it should be easier to study what malware do instead of what they contain. In this article, we propose to unmask Android malware hidden among benign applications using the observed information flows at the OS level. For achieving such a goal, we introduce a simple characterization of all the accountable information flows of a standard benign application. With such a model for benign apps, we lead some experiments evidencing that malware present some deviations from the expected normal behavior. Experiments show that our model recognizes most of the 3206 tested benign applications and spots most of the tested sophisticated malware (ransomware, rootkits, bootkit).