Security analysts widely use dynamic malware analysis environments to exercise malware samples and derive virus signatures. Unfortunately, malware authors are becoming more aware of such analysis environments. Therefore, many have embedded evasive logic into malware to probe execution environments before exposing malicious behaviors. Consequently, such analysis environments become useless and evasive malware can damage victim systems with unforeseen malicious activities. However, adopting evasive techniques to bypass dynamic malware analysis is a double-edged sword. While evasive techniques can avoid early detection through sandbox analysis, it also significantly constrains the spectrum of execution environments where the malware activates. In this paper, we exploit this dilemma and seek to reverse the challenge by camouflaging end-user execution environments into analysis-like environments using a lightweight deception engine called SCARECROW. We thoroughly evaluate SCARECROW with real evasive malware samples and demonstrate that we can successfully deactivate 89.56% of evasive malware samples and the variants of ransomware (e.g., WannaCry and Locky) with little or no impact on the most commonly used benign software. Our evaluation also shows that SCARECROW is able to steer state-of-the-art analysis environment fingerprinting techniques so that end-user execution environments with SCARECROW and malware analysis environments with SCARECROW become indistinguishable.

Distributed collaborative learning (DCL) paradigms enable building joint machine learning models from distrusted multi-party participants. Data confidentiality is guaranteed by retaining private training data on each participant's local infrastructure. However, this approach makes today's DCL design fundamentally vulnerable to data poisoning and backdoor attacks. It limits DCL's model accountability, which is key to backtracking problematic training data instances and their responsible contributors. In this paper, we introduce CALTRAIN, a centralized collaborative learning system that simultaneously achieves data confidentiality and model accountability. CALTRAIN enforces isolated computation via secure enclaves on centrally aggregated training data to guarantee data confidentiality. To support building accountable learning models, we securely maintain the links between training instances and their contributors. Our evaluation shows that the models generated by CALTRAIN can achieve the same prediction accuracy when compared to the models trained in non-protected environments. We also demonstrate that when malicious training participants tend to implant backdoors during model training, CALTRAIN can accurately and precisely discover the poisoned or mislabeled training data that lead to the runtime mispredictions.

Deep learning technologies, which are the key components of state-of-the-art Artificial Intelligence (AI) services, have shown great success in providing human-level capabilities for a variety of tasks, such as visual analysis, speech recognition, and natural language processing and etc. Building a production-level deep learning model is a non-trivial task, which requires a large amount of training data, powerful computing resources, and human expertises. Therefore, illegitimate reproducing, distribution, and the derivation of proprietary deep learning models can lead to copyright infringement and economic harm to model creators. Therefore, it is essential to devise a technique to protect the intellectual property of deep learning models and enable external verification of the model ownership.

The prevalence of malware in Android marketplaces is a growing and significant problem. Most existing studies focus on detecting Android malware or designing new security extensions to defend against specific types of attacks. In this paper, we perform an empirical study on analyzing the market-level and network-level behaviors of the Android malware ecosystem. We focus on studying whether there are interesting characteristics of those market accounts that distribute malware and specific networks that are mainly utilized by Android malware authors. We further investigate community patterns among Android malware from the perspective of their market account infrastructure and remote server infrastructure. Spurred by these analysis, we design a novel community inference algorithm to find more malicious apps by exploiting their community relationships. By using a small seed set (50) of known malicious apps, we can effectively find another extra 20 times of malicious apps, while maintaining considerable accuracy higher than 94%.

Nowadays, cyber criminals often build web infrastructures rather than a single server to conduct their malicious activities. In order to continue their malevolent activities without being detected, cyber criminals make efforts to conceal the core servers (e.g., C&C servers, exploit servers, and drop-zone servers) in the malicious web infrastructure. Such deliberate invisibility of those concealed malicious servers, however, makes them particularly distinguishable from benign web servers that are usually promoted to be public. In this paper, we conduct the first large-scale measurement study to investigate the visibility of both malicious and benign servers. From our intensive analysis of over 100,000 benign servers, 45,000 malicious servers and 40,000 redirections, we identify a set of distinct features of malicious web infrastructures from their locations, structures, roles, and relationships perspectives, and propose a lightweight yet effective detection system called VisHunter. VisHunter identifies malicious redirections from visible servers to invisible servers at the entryway of malicious web infrastructures. We evaluate VisHunter on both online public data and large-scale enterprise network traffic, and demonstrate that VisHunter can achieve an average 96.2% detection rate with only 0.9% false positive rate on the real enterprise network traffic.

HTTP is a popular channel for malware to communicate with malicious servers (e.g., Command & Control, drive-by download, drop-zone), as well as to attack benign servers. By utilizing HTTP requests, malware easily disguises itself under a large amount of benign HTTP traffic. Thus, identifying malicious HTTP activities is challenging. We leverage an insight that cyber criminals are increasingly using dynamic malicious infrastructures with multiple servers to be efficient and anonymous in (i) malware distribution (using redirectors and exploit servers), (ii) control (using C&C servers) and (iii) monetization (using payment servers), and (iv) being robust against server takedowns (using multiple backups for each type of servers). Instead of focusing on detecting individual malicious domains, we propose a complementary approach to identify a group of closely related servers that are potentially involved in the same malware campaign, which we term as Associated Server Herd (ASH). Our solution, SMASH (Systematic Mining of Associated Server Herds), utilizes an unsupervised framework to infer malware ASHs by systematically mining the relations among all servers from multiple dimensions. We build a prototype system of SMASH and evaluate it with traces from a large ISP. The result shows that SMASH successfully infers a large number of previously undetected malicious servers and possible zero-day attacks, with low false positives. We believe the inferred ASHs provide a better global view of the attack campaign that may not be easily captured by detecting only individual servers.

In this paper, through reverse engineering Twitter spammers' tastes (their preferred targets to spam), we aim at providing guidelines for building more effective social honeypots, and generating new insights to defend against social spammers. Specifically, we first perform a measurement study by deploying "benchmark" social honeypots on Twitter with diverse and fine-grained social behavior patterns to trap spammers. After five months' data collection, we make a deep analysis on how Twitter spammers find their targets. Based on the analysis, we evaluate our new guidelines for building effective social honeypots by implementing "advanced" honeypots. Particularly, within the same time period, using those advanced honeypots can trap spammers around 26 times faster than using "traditional" honeypots.

Malware often contains many system-resource-sensitive condition checks to avoid any duplicate infection, make sure to obtain required resources, or try to infect only targeted computers, etc. If we are able to extract the system resource constraints from malware code, and manipulate the environment state as vaccines, we would then be able to immunize a computer from infections. Towards this end, this paper provides the first systematic study and presents a prototype system, AUTOVAC, for automatically extracting the system resource constraints from malware code and generating vaccines based on the system resource conditions. Specifically, through monitoring the data propagation from system-resource-related system calls, AUTOVAC automatically identifies the environment related state of a computer. Through analyzing the environment state, AUTOVAC automatically generates vaccines. Such vaccines can be then injected into other computers, thereby being immune from future infections from the same malware or its polymorphic variants. We have evaluated AUTOVAC on a large set of real-world malware samples and successfully extracted working vaccines for many families including high-profile Conficker, Sality and Zeus. We believe AUTOVAC represents an appealing technique to complement existing malware defenses.

Inspired by the biological vaccines, we explore the possibility of developing similar vaccines for malware immunization. We provide the first systematic study towards this direction and present a prototype system, AGAMI, for automatic generation of vaccines for malware immunization. With a novel use of several dynamic malware analysis techniques, we show that it is possible to extract a lightweight vaccine from current malware, and after injecting such vaccine on clean machines, they can be immune from future infection from the same malware family. We evaluate AGAMI on a large set of real-world malware samples and successfully extract working vaccines for many families such as Conficker and Zeus. We believe it is an appealing complementary technique to existing malware defense solutions.

In this paper, we perform an empirical analysis of the cyber criminal ecosystem on Twitter. Essentially, through analyzing inner social relationships in the criminal account community, we find that criminal accounts tend to be socially connected, forming a small-world network. We also find that criminal hubs, sitting in the center of the social graph, are more inclined to follow criminal accounts. Through analyzing outer social relationships between criminal accounts and their social friends outside the criminal account community, we reveal three categories of accounts that have close friendships with criminal accounts. Through these analyses, we provide a novel and effective criminal account inference algorithm by exploiting criminal accounts' social relationships and semantic coordinations.