Search

In this paper we describe the SOREL-20M (Sophos/ReversingLabs-20 Million) dataset: a large-scale dataset consisting of nearly 20 million files with pre-extracted features and metadata, high-quality labels derived from multiple sources, information about vendor detections of the malware samples at the time of collection, and additional ``tags'' related to each malware sample to serve as additional targets. In addition to features and metadata, we also provide approximately 10 million ``disarmed'' malware samples -- samples with both the optional\_headers.subsystem and file\_header.machine flags set to zero -- that may be used for further exploration of features and detection strategies. We also provide Python code to interact with the data and features, as well as baseline neural network and gradient boosted decision tree models and their results, with full training and evaluation code, to serve as a starting point for further experimentation., Comment: Associated code available at: https://github.com/sophos-ai/SOREL-20M

Targeted phishing emails are on the rise and facilitate the theft of billions of dollars from organizations a year. While malicious signals from attached files or malicious URLs in emails can be detected by conventional malware signatures or machine learning technologies, it is challenging to identify hand-crafted social engineering emails which don't contain any malicious code and don't share word choices with known attacks. To tackle this problem, we fine-tune a pre-trained BERT model by replacing the half of Transformer blocks with simple adapters to efficiently learn sophisticated representations of the syntax and semantics of the natural language. Our Context-Aware network also learns the context representations between email's content and context features from email headers. Our CatBERT(Context-Aware Tiny Bert) achieves a 87% detection rate as compared to DistilBERT, LSTM, and logistic regression baselines which achieve 83%, 79%, and 54% detection rates at false positive rates of 1%, respectively. Our model is also faster than competing transformer approaches and is resilient to adversarial attacks which deliberately replace keywords with typos or synonyms.

Malware detection is a popular application of Machine Learning for Information Security (ML-Sec), in which an ML classifier is trained to predict whether a given file is malware or benignware. Parameters of this classifier are typically optimized such that outputs from the model over a set of input samples most closely match the samples' true malicious/benign (1/0) target labels. However, there are often a number of other sources of contextual metadata for each malware sample, beyond an aggregate malicious/benign label, including multiple labeling sources and malware type information (e.g., ransomware, trojan, etc.), which we can feed to the classifier as auxiliary prediction targets. In this work, we fit deep neural networks to multiple additional targets derived from metadata in a threat intelligence feed for Portable Executable (PE) malware and benignware, including a multi-source malicious/benign loss, a count loss on multi-source detections, and a semantic malware attribute tag loss. We find that incorporating multiple auxiliary loss terms yields a marked improvement in performance on the main detection task. We also demonstrate that these gains likely stem from a more informed neural network representation and are not due to a regularization artifact of multi-target learning. Our auxiliary loss architecture yields a significant reduction in detection error rate (false negatives) of 42.6% at a false positive rate (FPR) of $10^{-3}$ when compared to a similar model with only one target, and a decrease of 53.8% at $10^{-5}$ FPR., Comment: Pre-print of a manuscript submitted to Usenix Security Symposium 2019

We propose several generalized linear models (GLMs) to predict the number of successful cyber intrusions (or "intrusions") into an organization's computer network, where the rate at which intrusions occur is a function of the following observable characteristics of the organization: (i) domain name server (DNS) traffic classified by their top-level domains (TLDs); (ii) the number of network security policy violations; and (iii) a set of predictors that we collectively call "cyber footprint" that is comprised of the number of hosts on the organization's network, the organization's similarity to educational institution behavior (SEIB), and its number of records on scholar.google.com (ROSG). In addition, we evaluate the number of intrusions to determine whether these events follow a Poisson or negative binomial (NB) probability distribution. We reveal that the NB GLM provides the best fit model for the observed count data, number of intrusions per organization, because the NB model allows the variance of the count data to exceed the mean. We also show that there are restricted and simpler NB regression models that omit selected predictors and improve the goodness-of-fit of the NB GLM for the observed data. With our model simulations, we identify certain TLDs in the DNS traffic as having significant impact on the number of intrusions. In addition, we use the models and regression results to conclude that the number of network security policy violations are consistently predictive of the number of intrusions.

When the cost of misclassifying a sample is high, it is useful to have an accurate estimate of uncertainty in the prediction for that sample. There are also multiple types of uncertainty which are best estimated in different ways, for example, uncertainty that is intrinsic to the training set may be well-handled by a Bayesian approach, while uncertainty introduced by shifts between training and query distributions may be better-addressed by density/support estimation. In this paper, we examine three types of uncertainty: model capacity uncertainty, intrinsic data uncertainty, and open set uncertainty, and review techniques that have been derived to address each one. We then introduce a unified hierarchical model, which combines methods from Bayesian inference, invertible latent density inference, and discriminative classification in a single end-to-end deep neural network topology to yield efficient per-sample uncertainty estimation in a detection context. This approach addresses all three uncertainty types and can readily accommodate prior/base rates for binary detection. We then discuss how to extend this model to a more generic multiclass recognition context.

Malicious email attachments are a growing delivery vector for malware. While machine learning has been successfully applied to portable executable (PE) malware detection, we ask, can we extend similar approaches to detect malware across heterogeneous file types commonly found in email attachments? In this paper, we explore the feasibility of applying machine learning as a static countermeasure to detect several types of malicious email attachments including Microsoft Office documents and Zip archives. To this end, we collected a dataset of over 5 million malicious/benign Microsoft Office documents from VirusTotal for evaluation as well as a dataset of benign Microsoft Office documents from the Common Crawl corpus, which we use to provide more realistic estimates of thresholds for false positive rates on in-the-wild data. We also collected a dataset of approximately 500k malicious/benign Zip archives, which we scraped using the VirusTotal service, on which we performed a separate evaluation. We analyze predictive performance of several classifiers on each of the VirusTotal datasets using a 70/30 train/test split on first seen time, evaluating feature and classifier types that have been applied successfully in commercial antimalware products and R&D contexts. Using deep neural networks and gradient boosted decision trees, we are able to obtain ROC curves with > 0.99 AUC on both Microsoft Office document and Zip archive datasets. Discussion of deployment viability in various antimalware contexts is provided., Comment: Pre-print of a manuscript submitted to IEEE Symposium on Technologies for Homeland Security (HST)

Malicious web content is a serious problem on the Internet today. In this paper we propose a deep learning approach to detecting malevolent web pages. While past work on web content detection has relied on syntactic parsing or on emulation of HTML and Javascript to extract features, our approach operates directly on a language-agnostic stream of tokens extracted directly from static HTML files with a simple regular expression. This makes it fast enough to operate in high-frequency data contexts like firewalls and web proxies, and allows it to avoid the attack surface exposure of complex parsing and emulation code. Unlike well-known approaches such as bag-of-words models, which ignore spatial information, our neural network examines content at hierarchical spatial scales, allowing our model to capture locality and yielding superior accuracy compared to bag-of-words baselines. Our proposed architecture achieves a 97.5% detection rate at a 0.1% false positive rate, and classifies small-batched web pages at a rate of over 100 per second on commodity hardware. The speed and accuracy of our approach makes it appropriate for deployment to endpoints, firewalls, and web proxies.

We analyze sets of intrusion detection records observed on the networks of several large, nonresidential organizations protected by a form of intrusion detection and prevention service. Our analyses reveal that the process of intrusion detection in these networks exhibits a significant degree of burstiness as well as strong memory, with burstiness and memory properties that are comparable to those of natural processes driven by threshold effects, but different from bursty human activities. We explore time-series models of these observable network security incidents based on partially observed data using a hidden Markov model with restricted hidden states, which we fit using Markov Chain Monte Carlo techniques. We examine the output of the fitted model with respect to its statistical properties and demonstrate that the model adequately accounts for intrinsic "bursting" within observed network incidents as a result of alternation between two or more stochastic processes. While our analysis does not lead directly to new detection capabilities, the practical implications of gaining better understanding of the observed burstiness are significant, and include opportunities for quantifying a network's risks and defensive efforts., Comment: This is a version of the paper that is to appear as Harang, R., & Kott, A. (2017). Burstiness of Intrusion Detection Process: Empirical Evidence and a Modeling Approach. IEEE Transactions on Information Forensics and Security

The high mobility of Army tactical networks, combined with their close proximity to hostile actors, elevates the risks associated with short-range network attacks. The connectivity model for such short range connections under active operations is extremely fluid, and highly dependent upon the physical space within which the element is operating, as well as the patterns of movement within that space. To handle these dependencies, we introduce the notion of "key cyber-physical terrain": locations within an area of operations that allow for effective control over the spread of proximity-dependent malware in a mobile tactical network, even as the elements of that network are in constant motion with an unpredictable pattern of node-to-node connectivity. We provide an analysis of movement models and approximation strategies for finding such critical nodes, and demonstrate via simulation that we can identify such key cyber-physical terrain quickly and effectively., Comment: 16 pages, extended version of paper published in Proceedings of the International Workshop on Security & Privacy Analytics (IWSPA)

Program authorship attribution has implications for the privacy of programmers who wish to contribute code anonymously. While previous work has shown that complete files that are individually authored can be attributed, we show here for the first time that accounts belonging to open source contributors containing short, incomplete, and typically uncompilable fragments can also be effectively attributed. We propose a technique for authorship attribution of contributor accounts containing small source code samples, such as those that can be obtained from version control systems or other direct comparison of sequential versions. We show that while application of previous methods to individual small source code samples yields an accuracy of about 73% for 106 programmers as a baseline, by ensembling and averaging the classification probabilities of a sufficiently large set of samples belonging to the same author we achieve 99% accuracy for assigning the set of samples to the correct author. Through these results, we demonstrate that attribution is an important threat to privacy for programmers even in real-world collaborative environments such as GitHub. Additionally, we propose the use of calibration curves to identify samples by unknown and previously unencountered authors in the open world setting. We show that we can also use these calibration curves in the case that we do not have linking information and thus are forced to classify individual samples directly. This is because the calibration curves allow us to identify which samples are more likely to have been correctly attributed. Using such a curve can help an analyst choose a cut-off point which will prevent most misclassifications, at the cost of causing the rejection of some of the more dubious correct attributions.

Machine learning models are frequently used to solve complex security problems, as well as to make decisions in sensitive situations like guiding autonomous vehicles or predicting financial market behaviors. Previous efforts have shown that numerous machine learning models were vulnerable to adversarial manipulations of their inputs taking the form of adversarial samples. Such inputs are crafted by adding carefully selected perturbations to legitimate inputs so as to force the machine learning model to misbehave, for instance by outputting a wrong class if the machine learning task of interest is classification. In fact, to the best of our knowledge, all previous work on adversarial samples crafting for neural network considered models used to solve classification tasks, most frequently in computer vision applications. In this paper, we contribute to the field of adversarial machine learning by investigating adversarial input sequences for recurrent neural networks processing sequential data. We show that the classes of algorithms introduced previously to craft adversarial samples misclassified by feed-forward neural networks can be adapted to recurrent neural networks. In a experiment, we show that adversaries can craft adversarial sequences misleading both categorical and sequential recurrent neural networks.

The ability to identify authors of computer programs based on their coding style is a direct threat to the privacy and anonymity of programmers. While recent work found that source code can be attributed to authors with high accuracy, attribution of executable binaries appears to be much more difficult. Many distinguishing features present in source code, e.g. variable names, are removed in the compilation process, and compiler optimization may alter the structure of a program, further obscuring features that are known to be useful in determining authorship. We examine programmer de-anonymization from the standpoint of machine learning, using a novel set of features that include ones obtained by decompiling the executable binary to source code. We adapt a powerful set of techniques from the domain of source code authorship attribution along with stylistic representations embedded in assembly, resulting in successful de-anonymization of a large set of programmers. We evaluate our approach on data from the Google Code Jam, obtaining attribution accuracy of up to 96% with 100 and 83% with 600 candidate programmers. We present an executable binary authorship attribution approach, for the first time, that is robust to basic obfuscations, a range of compiler optimization settings, and binaries that have been stripped of their symbol tables. We perform programmer de-anonymization using both obfuscated binaries, and real-world code found "in the wild" in single-author GitHub repositories and the recently leaked Nulled.IO hacker forum. We show that programmers who would like to remain anonymous need to take extreme countermeasures to protect their privacy., Comment: 15 pages