Distributed detection of anomalous internet sessions

Financial service providers are moving many services online reducing their costs and facilitating customers¿ interaction. Unfortunately criminals have quickly found several ways to avoid most security measures applied to browsers and banking sites. The use of highly dangerous malware has become the most significant threat and traditional signature-detection methods are nowadays easily circumvented due to the amount of new samples and the use of sophisticated evasion techniques. Antivirus vendors and malware experts are pushed to seek for new methodologies to improve the identification and understanding of malicious applications behavior and their targets. Financial institutions are now playing an important role by deploying their own detection tools against malware that specifically affect their customers. However, most detection approaches tend to base on sequence of bytes in order to create new signatures. This thesis approach is based on new sources of information: the web logs generated from each banking session, the normal browser execution and customers mobile phone behavior. The thesis can be divided in four parts: The first part involves the introduction of the thesis along with the presentation of the problems and the methodology used to perform the experimentation. The second part describes our contributions to the research, which are based in two areas: *Server side: Weblogs analysis. We first focus on the real time detection of anomalies through the analysis of web logs and the challenges introduced due to the amount of information generated daily. We propose different techniques to detect multiple threats by deploying per user and global models in a graph based environment that will allow increase performance of a set of highly related data. *Customer side: Browser analysis. We deal with the detection of malicious behaviors from the other side of a banking session: the browser. Malware samples must interact with the browser in order to retrieve or add information. Such relation interferes with the normal behavior of the browser. We propose to develop models capable of detecting unusual patterns of function calls in order to detect if a given sample is targeting an specific financial entity. In the third part, we propose to adapt our approaches to mobile phones and Critical Infrastructures environments. The latest online banking attack techniques circumvent protection schemes such password verification systems send via SMS. Man in the Mobile attacks are capable of compromising mobile devices and gaining access to SMS traffic. Once the Transaction Authentication Number is obtained, criminals are free to make fraudulent transfers. We propose to model the behavior of the applications related messaging services to automatically detect suspicious actions. Real time detection of unwanted SMS forwarding can improve the effectiveness of second channel authentication and build on detection techniques applied to browsers and Web servers. Finally, we describe possible adaptations of our techniques to another area outside the scope of online banking: critical infrastructures, an environment with similar features since the applications involved can also be profiled. Just as financial entities, critical infrastructures are experiencing an increase in the number of cyber attacks, but the sophistication of the malware samples utilized forces to new detection approaches. The aim of the last proposal is to demonstrate the validity of out approach in different scenarios. Conclusions. Finally, we conclude with a summary of our findings and the directions for future work