Web sites are often used for diffusing malware; an increasingly number of attacks are performed by delivering malicious code in web pages: drive-by download, malvertisement, rogueware, phishing are just the most common examples. In this scenario, JavaScript plays an important role, as it allows to insert code into the web page that will be executed on the client machine, letting the attacker to perform a plethora of actions which are necessary to successfully accomplish an attack. Existing techniques for detecting malicious JavaScript suffer from some limitations like: the capability of recognizing only known attacks, being tailored only to specific attacks, or being ineffective when appropriate evasion techniques are implemented by attackers. In this paper we propose to use system calls to detect malicious JavaScript. The main advantage is that capturing the system calls allows a description of the attack at a very high level of abstraction. On the one hand, this limits the evasion techniques which could succeed, and, on the other hand, produces a very high detection accuracy (96%), as experimentation demonstrated.
