Patent Grant
9398047
Embodiments are related to the detection of phishing Universal Resource Locators (URLs) delivered through electronic messages such as email. Phishing detection refers to the detection of URLs in, for example, emails that purport to be from a legitimate and trustworthy source but that, in fact, do not. Such phishing URLs often are used in attempts to collect personal and financial information from the unsuspecting recipient, often for unauthorized purposes.
The goal of the phisher is most often to capture critical data such as credit card number or login/password credentials. For this purpose, the phisher sends an email to the victim that contains a URL that will lead the victim to a forged website where the victim is induced to enter the sought-after personal and financial information.
The user experience is specific to each brand. In order to maximize the capture of critical data in a forged website, the user experience occasioned by viewing and interacting with the phishing email and with the forged website should to be as close as possible to the genuine user experience with a legitimate email and website. For example, a phishing email received by the victim often contain text and graphics—typically, a known and familiar brand logo—to convince the victim to click on a URL link of the forged website and enter his or her credentials therein. Toward that end, the forged website URL often contain keywords that are close to the genuine website URL and the forged website often contains text, style sheets, graphics and user experience that resemble those of the genuine website.
One embodiment is a method of determining whether a URL is a phishing URL through real-time exploration and analysis that carry out a number of determinations that, in the aggregate, determine the likelihood that a received URL is a phishing URL, as is URL 108 in
The exploration of the URL, as shown at B206, may comprise comparing the URL or a portion or portions thereof with a database (the same or a different database than referred to above) of phishing signatures. Such phishing signatures may comprise, for example, a list of regular expressions that are most often associated with phishing attempts. Such comparison may comprise comparing the content of the webpage pointed to by the URL under consideration with database records of known phishing webpages signatures. A match of such a comparison may result, according to one embodiment, with a determination that the URL is a phishing URL. If no match is found, the method may proceed to block B207. It to be noted, however, that blocks B201-B206 may be carried out in an order that is different than that shown in
At B207, the URL (which thus far has resisted attempts to classify it as a phishing URL or as a non-phishing URL in previous determinations) may be submitted to a phishing probability engine, the output of which may be interpreted as a probability that the submitted URL under consideration is, in fact, a phishing URL. The probability may be expressed numerically, or may be expressed as a more user-friendly phishing probability rating. For example, the output of the phishing probability engine may comprise ratings such as “Most Likely Not a Phishing URL”, “Somewhat Probable Phishing URL” or “Most Likely a Phishing URL” or functionally equivalent ratings with a lesser or greater degree of granularity. According to one embodiment, the phishing probability engine may comprise supervised learning models and associated algorithms to analyze data and recognize patterns. One embodiment utilizes a Support Vector Machine (SVM) classifier on the URL itself and the webpage content.
There are a great many well-known brands and each of these brands has separate characteristics, color and font scheme and look and feel. Examples of such brands include, for example, Microsoft, PayPal, Apple or Bank of America. Well-known brands with which users interact frequently are prime candidates for phishing attacks. Rather than extracting features that are common to all brands, one embodiment comprises and accesses a knowledge database of brands configured to enable the present system to extract therefrom items that may be characteristic or specific to each brand.
Brand Elements
According to one embodiment, a brand is identified by a unique name such as Apple, PayPal, Bank of America, Chase or Yahoo. A brand contains a list of elements that defines the knowledge base relative to this brand. According to one embodiment, a knowledge database of brands configured to enable extraction therefrom of items that are characteristic or specific to each brand may include one or more of the following elements:
According to one embodiment, a brand may be defined as a logical construct that includes several elements. Such a logical construct, according to one embodiment, may be implemented as a document type definition (DTD). Other logical constructs may be devised. A DTD is a set of markup declarations that define a document type for an SGML-family markup language (SGML, XML, HTML) and defines the legal building blocks of an XML document. A DTD defines the document structure with a list of legal elements and attributes. A DID that encapsulates a brand, according to one embodiment, may be implemented as an XML file having the following form:
The following is an exemplary brand description for the Chase bank brand:
The following is an exemplary brand description for the Apple brand:
Vector Definition
In order to classify a URL as being a legitimate or a suspected phishing URL, one embodiment computes a vector that is suitable to be input to the phishing probability detection engine. One embodiment computes a multi-dimensional vector of binary values, either 0 or 1. One implementation computes a 14-dimensional vector of binary values. Such a vector may be represented by, for example, a 14 bits array. Each dimension (represented by one bit) represents a feature: the bit is set to 1 if the feature condition is met, otherwise the bit is set to 0.
The features of one implementation are shown below, according to one embodiment.
As shown in the table below, some of these features are brand-dependent and rely on a brand selection process that will be described further. In the table below, those features having an “X” in the Brand Dependent column are brand-dependent.
As shown in
At B503, it may be determine whether the determination of the selected features vector features above is sufficient to enable an identification of the brand that is the subject of the phishing attempt (if such phishing attempt exists). According to one embodiment, the identification of the brand may be carried out according to the method shown and described relative to
After the determination of the value of brand-specific phishing features or after it is determined in B504 that the specific brand may not be identified from the examined features, block B505 may be carried out, to determine the value of remaining, non-brand-specific phishing features such as, for example:
The resultant features vector may now be input to the phishing probability engine, as shown at 506.
A brand identification algorithm according to one embodiment is shown in
The following phishing URL example uses the Chase brand name, for exemplary purposes only.
Examination of this phishing URL, according to one embodiment, would lead to a brand identification of Chase, as Chase is a keyword element matching URL path element at B602 in
This phishing link leads to a brand identification of Apple as itunes is a keyword element matching URL subdomain element at B601 in
Compute Phishing Probability with SVM Classifier
The computed input vector may now be input to phishing probability engine. According to one embodiment, the phishing probability engine may comprise a Support Vector Machine (SVM) classifier. One embodiment of the phishing probability engine uses a binary SVM classifier, in which the two classes N and P are
Herein, an element is a pair of two files. According to one embodiment, the first file of the pair of files of the element is a URL file, containing the URL under investigation. The second file of the pair of files of the element is an HTML file containing the webpage pointed to by the URL. According to one implementation, the filename of the first file is a hash of, for example, a quantity such as the current timestamp and the URL under investigation. The extension of the first file may be, for example, “.url”. Similarly, the filename of the second file may be a hash of, for example, a quantity such as the current timestamp and the content of the webpage pointed to by the link (e.g., URL) in the email. The extension of the second file may be, for example, “.html”. According to one embodiment, the hash may be a message digest algorithm such as an MD5 hash, although other hashes may be utilized as well. For example, the two files may be named as follows:
To train the SVM classifier, it may be provided with a corpus of P (phishing elements) and N (non-phishing) elements. This corpus may be updated periodically as new phishing attempts are discovered, to follow the phishing trend. The training and testing of the SVM classifier produces a SVM model that may be used by the phishing probability engine.
According to one embodiment, for an input vector V (e.g., the 14-dimensional input vector discussed herein), the SVM classifier of the phishing probability engine produces a probability: the probability that input vector V belongs to the P class, the class of phishing elements. This probability may then be used to decide whether the URL under investigation is likely a phishing URL. Subsequently, actions such as deleting, guaranteeing or placing an email in a “Junk” folder, may be carried out, based upon the computed probability.
Embodiments of the present invention are related to the use of computing device 712, 708 to detect and compute a probability that received email contains a phishing URL. According to one embodiment, the methods and systems described herein may be provided by one or more computing devices 712, 708 in response to processor(s) 802 executing sequences of instructions contained in memory 804. Such instructions may be read into memory 804 from another computer-readable medium, such as data storage device 807. Execution of the sequences of instructions contained in memory 804 causes processor(s) 802 to perform the steps and have the functionality described herein. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions to implement the described embodiments. Thus, embodiments are not limited to any specific combination of hardware circuitry and software. Indeed, it should be understood by those skilled in the art that any suitable computer system may implement the functionality described herein. The computing devices may include one or a plurality of microprocessors working to perform the desired functions. In one embodiment, the instructions executed by the microprocessor or microprocessors are operable to cause the microprocessor(s) to perform the steps described herein. The instructions may be stored in any computer-readable medium. In one embodiment, they may be stored on a non-volatile semiconductor memory external to the microprocessor, or integrated with the microprocessor. In another embodiment, the instructions may be stored on a disk and read into a volatile semiconductor memory before execution by the microprocessor.
While certain embodiments of the disclosure have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the disclosure. Indeed, the novel methods, devices and systems described herein may be embodied in a variety of other forms. Furthermore, various omissions, substitutions and changes in the form of the methods and systems described herein may be made without departing from the spirit of the disclosure. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the scope and spirit of the disclosure. For example, those skilled in the art will appreciate that in various embodiments, the actual physical and logical structures may differ from those shown in the figures. Depending on the embodiment, certain steps described in the example above may be removed, others may be added. Also, the features and attributes of the specific embodiments disclosed above may be combined in different ways to form additional embodiments, all of which fall within the scope of the present disclosure. Although the present disclosure provides certain preferred embodiments and applications, other embodiments that are apparent to those of ordinary skill in the art, including embodiments which do not provide all of the features and advantages set forth herein, are also within the scope of this disclosure. Accordingly, the scope of the present disclosure is intended to be defined only by reference to the appended claims.
| Number | Name | Date | Kind |
|---|---|---|---|
| 7422115 | Zager et al. | Sep 2008 | B2 |
| 7634810 | Goodman et al. | Dec 2009 | B2 |
| 7668921 | Proux | Feb 2010 | B2 |
| 8015598 | Hird | Sep 2011 | B2 |
| 8032594 | Helsper et al. | Oct 2011 | B2 |
| 8079087 | Spies | Dec 2011 | B1 |
| 8135790 | Castelli | Mar 2012 | B1 |
| 8141150 | Krishnamurthy | Mar 2012 | B1 |
| 8209381 | Sinn et al. | Jun 2012 | B2 |
| 8220047 | Soghoian et al. | Jul 2012 | B1 |
| 8291065 | Goodman et al. | Oct 2012 | B2 |
| 8438462 | Feng et al. | May 2013 | B2 |
| 8468597 | Warner | Jun 2013 | B1 |
| 8516581 | Hsu et al. | Aug 2013 | B2 |
| 8528079 | Wang | Sep 2013 | B2 |
| 8566938 | Prakash et al. | Oct 2013 | B1 |
| 8601586 | Boutros et al. | Dec 2013 | B1 |
| 8621614 | Vaithilingam et al. | Dec 2013 | B2 |
| 8713677 | Soghoian et al. | Apr 2014 | B2 |
| 8839369 | Dai et al. | Sep 2014 | B1 |
| 8839418 | Hulten et al. | Sep 2014 | B2 |
| 8874658 | Khalsa et al. | Oct 2014 | B1 |
| 20050086161 | Gallant | Apr 2005 | A1 |
| 20050228899 | Wendkos et al. | Oct 2005 | A1 |
| 20060053293 | Zager et al. | Mar 2006 | A1 |
| 20060075027 | Zager et al. | Apr 2006 | A1 |
| 20060075028 | Zager et al. | Apr 2006 | A1 |
| 20060080735 | Brinson et al. | Apr 2006 | A1 |
| 20060095955 | Vong et al. | May 2006 | A1 |
| 20060101120 | Helsper et al. | May 2006 | A1 |
| 20060117307 | Averbuch | Jun 2006 | A1 |
| 20060123464 | Goodman et al. | Jun 2006 | A1 |
| 20060168066 | Helsper et al. | Jul 2006 | A1 |
| 20070033639 | Goodman et al. | Feb 2007 | A1 |
| 20070039038 | Goodman et al. | Feb 2007 | A1 |
| 20070078936 | Quinlan | Apr 2007 | A1 |
| 20070094500 | Shannon et al. | Apr 2007 | A1 |
| 20070136806 | Berman | Jun 2007 | A1 |
| 20070162366 | Tanaka et al. | Jul 2007 | A1 |
| 20070192855 | Hulten et al. | Aug 2007 | A1 |
| 20070283000 | Proux | Dec 2007 | A1 |
| 20080141342 | Curnyn | Jun 2008 | A1 |
| 20080177841 | Sinn et al. | Jul 2008 | A1 |
| 20080244715 | Pedone | Oct 2008 | A1 |
| 20090077383 | de Monseignat et al. | Mar 2009 | A1 |
| 20090077637 | Santos et al. | Mar 2009 | A1 |
| 20090089859 | Cook et al. | Apr 2009 | A1 |
| 20090138950 | Hird | May 2009 | A1 |
| 20090216795 | Cohen et al. | Aug 2009 | A1 |
| 20090328208 | Peters | Dec 2009 | A1 |
| 20100042687 | Wang | Feb 2010 | A1 |
| 20100043071 | Wang | Feb 2010 | A1 |
| 20100251380 | Zhang | Sep 2010 | A1 |
| 20100306845 | Vaithilingam et al. | Dec 2010 | A1 |
| 20100313266 | Feng | Dec 2010 | A1 |
| 20110035317 | Carlson et al. | Feb 2011 | A1 |
| 20120158626 | Zhu | Jun 2012 | A1 |
| 20120259933 | Bardsley | Oct 2012 | A1 |
| 20130086677 | Ma | Apr 2013 | A1 |
| 20130145462 | Hsu et al. | Jun 2013 | A1 |
| 20130238721 | Patel | Sep 2013 | A1 |
| 20140041024 | Larkins | Feb 2014 | A1 |
| 20140082521 | Carolan et al. | Mar 2014 | A1 |
| 20140298460 | Xue | Oct 2014 | A1 |
| Entry |
|---|
| International Search Report and Written Opinion of the International Searching Authority dated Mar. 11, 2016 in PCT/US2016/012285. |
| What is Safe Browsing? Google Developers, downloaded from https://developers.google.com/safe-browsing/ on Nov. 15, 2014. |
| Wikipedia, Google Safe Browsing, downloaded from http://en.wildpedia.org./wiki/Google—Safe—Browsing on Nov. 15, 2014. |
| Large-Scale Automatic Classification of Phishing pages, Whittaker, Colin, Ryner, Briand adn Nazif, Marria, downloaded from https://www.isoc.org/isoc/conferences/ndss/10/pdf/08.pdf on Nov. 15, 2014. |
| SmartScreen Filter: frequently asked questions—Windows Help, downloaded from http://windows.microsoft.com/en-us/windows/smartscreen-filter-faq#1TC=windows-7 on Nov. 15, 2014. |
| Yandex Safe Browsing API, downloaded from http://api.yandex.com/safebrowsing/ on Nov. 15, 2014. |
| Number | Date | Country | |
|---|---|---|---|
| 20160142439 A1 | May 2016 | US |
