Palo Alto Networks, Inc.

Yuchen Zhou

Oleksii Starov

Fang Liu

William Redington Hewlett, Ii

Brody James Kutt

Jun Wang

Wei Xu

Xiao Meng Zhang

Zhanhao Chen

Michael Edward Weber

Techniques for providing innocent until proven guilty (IUPG) solutions for building and using adversary resistant and false positive resistant deep learning models are disclosed. In some embodiments, a system, process, and/or computer program product includes storing a set comprising one or more innocent until proven guilty (IUPG) models for static analysis of a sample; performing a static analysis of content associated with the sample, wherein performing the static analysis includes using at least one stored IUPG model; and determining that the sample is malicious based at least in part on the static analysis of the content associated with the sample, and in response to determining that the sample is malicious, performing an action based on a security policy.

Innocent until proven guilty (IUPG): adversary resistant and false positive resistant deep learning models

The technology presented herein enables the use of a clustering algorithm to identify additional malicious domains based on known malicious domains. A domain identifier system identifies a first plurality of domain names associated with a malicious domain campaign and seeding a first clustering algorithm with the first plurality of domain names. After seeding the first clustering algorithm, the domain identifier system uses the first clustering algorithm to process passive domain name system (DNS) records to identify and group a second plurality of domain names associated with the malicious domain campaign.

Identification of malicious domain campaigns using unsupervised clustering

Web sites are crawled using multiple browser profiles to avoid malicious cloaking. Based on web page content returned from HTTP requests using the multiple browser profiles, web sites returning substantively different content to HTTP requests for different browser profiles are identified. Web sites are further filtered by common cloaking behavior, and redirect scripts are extracted from web page content that performed cloaking. Signatures comprising tokenized versions of the redirect scripts are generated and compared to a database of known cloaking signatures. URLs corresponding to signatures having approximate matches with signatures in the database are flagged for recrawling. Recrawled URLs are verified for malicious cloaking again using HTTP requests from multiple browser profiles.

Intelligent signature-based anti-cloaking web recrawling

To perform pattern-based detection of malicious URLs, patterns are first generated from known URLs to build a pattern repository. A URL is first normalized and parsed, and keywords are extracted and stored in an additional repository of keywords. Tokens are then determined from the parsed URL and tags are associated with the parsed substrings. Substring text may also be replaced with general identifying information. Patterns generated from known malicious and benign URLs satisfying certain criteria are published to a pattern repository of which can be accessed during subsequent detection operations. During detection, upon identifying a request which indicates an unknown URL, the URL is parsed and tokenized to generate a pattern. The repository of malicious URL patterns is queried to determine if a matching malicious URL pattern can be identified. If a matching malicious URL pattern is identified, the URL is detected as malicious.

Pattern-based malicious URL detection

A system/process/computer program product for building multi-representational learning models for static analysis of source code includes receiving training data, wherein the training data includes a set of source code files for training a multi-representational learning (MRL) model for classifying malicious source code and benign source code based on a static analysis; generating a first feature vector based on a set of characters extracted from the set of source code files; generating a second feature vector based on a set of tokens extracted from the set of source code files; and performing an ensemble of the first feature vector and the second feature vector to form a target feature vector for classifying malicious source code and benign source code based on the static analysis.

Building multi-representational learning models for static analysis of source code

Techniques for multi-representational learning models for static analysis of source code are disclosed. In some embodiments, a system/process/computer program product for multi-representational learning models for static analysis of source code includes receiving at a networked device a set comprising one or more multi-representation learning (MRL) models for static analysis of source code; performing a static analysis of source code associated with a sample received at the network device, wherein performing the static analysis includes using at least one MRL model; and determining that the sample is malicious based at least in part on the static analysis of the source code associated with the sample and without performing dynamic analysis of the sample, and in response to determining that the sample is malicious, performing an action based on a security policy.

Multi-representational learning models for static analysis of source code

An author of a malicious websites campaign (scam or phishing) likely uses a legitimate third-party service to facilitate the malicious campaign. An example includes legitimate CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart) services to conceal the malicious campaign from automated security scanners. A security service/platform can employ a detection pipeline that leverages use of CAPTCHA keys across websites of a malicious websites campaign. Websites that use CAPTCHA keys found in known malicious websites can at least be identified as suspect and communicated to firewalls.

Malicious website discovery using legitimate third party identifiers

Techniques for multi-representational learning models for static analysis of source code are disclosed. In some embodiments, a system/process/computer program product for multi-representational learning models for static analysis of source code includes storing on a networked device a set comprising one or more multi-representation learning (MRL) models for static analysis of source code; performing a static analysis of source code associated with a sample received at the network device, wherein performing the static analysis includes using at least one stored MRL model; and determining that the sample is malicious based at least in part on the static analysis of the source code associated with the received sample, and in response to determining that the sample is malicious, performing an action based on a security policy.

The technology presented herein enables malicious websites to be discovered based on web analytics identifiers associated therewith. In a particular embodiment, a method provides obtaining information about one or more first websites that are known to be malicious. The method further provides identifying one or more first analytics identifiers that are each associated with at least one of the first websites and identifying one or more second websites that are each associated with at least one of the first analytics identifiers.

Growing community of inventors

Newark, CA, United States of America

Yuchen Zhou