System and method for run-time object classification

Abstract

One embodiment of an electronic device comprises a processor and a memory accessible by the processor. The memory comprises virtual execution logic and run-time classifier logic. The virtual execution logic includes at least one virtual machine that is configured to virtually process content within an object under analysis and monitor for anomalous behaviors during the virtual processing that are indicative of malware. The run-time classifier logic performs, during run-time, a first analysis on the monitored anomalous behaviors and a pre-stored identifier to determine if the monitored anomalous behaviors indicate that the object is malware belonging to a classified malware family. The pre-stored identifier is a collection of data associated with anomalous behaviors that uniquely identify the malware family.

Description

FIELD

Embodiments of the disclosure relate to the field of data security. More specifically, one embodiment of the disclosure relates to a run-time classification of malicious objects, including advanced persistent threats (APTs).

GENERAL BACKGROUND

Over the last decade, malicious software (malware) has become a pervasive problem for Internet users. In some situations, malware is a program or file that is embedded within downloadable content and designed to adversely influence or attack normal operations of a computer. Examples of different types of malware may include bots, computer viruses, worms, Trojan horses, spyware, adware, or any other programming that operates within an electronic device (e.g., computer, smartphone, server, router, wearable technology, or other types of electronics with data processing capabilities) without permission by the user or an administrator.

In general, an advanced persistent threat (APT) is malware that targets an entity and may be configured to exfiltrate (send out) information that is accessible to that entity. The targeted entity may include an individual or organization with high value information (e.g., classified or sensitive defense secrets, trade secrets, intellectual property, or the like). Currently, the classification of different types of malware, such as APTs for example, is quite resource intensive. For APTs, classification may require off-line system and workforce training.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the invention are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings, in which like references indicate similar elements and in which:

FIG. 1 is an exemplary block diagram of an operational flow of a run-time malware classification conducted within an electronic device.

FIG. 2 is an exemplary block diagram of a first illustrative embodiment of an APT detection system that is adapted to perform run-time APT classification on incoming objects.

FIG. 3 is an exemplary block diagram of an illustrative embodiment of a communication system deploying run-time APT classification.

FIG. 4 is an illustrative embodiment of operations conducted by family identifier generator logic to generate a family identifier, such as an APT family identifier.

FIG. 5 is a detailed exemplary of a component diagram of the APT server according to one embodiment of the invention.

FIGS. 6A-6B are illustrative embodiments a method for identifying and classifying APT objects.

FIGS. 7A-7B are exemplary web-interfaces for submitting a suspected object to the APT server from the client device and receipt of a warning message.

FIG. 8 is an exemplary block diagram of a second illustrative embodiment of an electronic device that is adapted to perform run-time APT classification on incoming objects.

FIG. 9 is an illustrative embodiment of a communication system implemented with the electronic device FIG. 8 with run-time APT classification functionality.

FIG. 10 is an illustrative embodiment of a method for identifying malicious objects supplemented by use of a run-time classifier.

DETAILED DESCRIPTION

I. Overview

Unlike conventional Advanced Persistent Threat (APT) detection systems, a first embodiment of the disclosure are directed to an APT detection system that is capable of automatically and quickly identifying a suspect object based on previously classified APT families in order to enable a network administrator or user to more easily understand the severity, origin, or tendencies of the recently detected APT.

A second embodiment of the disclosure is directed to a networked security appliance that is deployed with logic (e.g. run-time classifier) to accelerate detection of APT and non-APT malicious objects based on anomalous behaviors uncovered during virtual processing of the suspect object and anomalous behaviors that uniquely identify both known APT families and other malware type families.

As generally stated, an “APT” is a type of malware that is directed at a targeted entity and seeks to surveil, extract, and/or manipulate data to which the targeted entity would have access. In some instances, in lieu of data gathering, APTs may seek to perform nation state attacks for the purposes of political terrorism or cyber/industrial espionage. Hence, APTs are generally viewed as more serious threats because these attacks target a specific person or persons to acquire information (normally for nefarious reasons) and are persistent. Herein, a number of benefits may be realized through classification of APT and malware families for subsequent analysis, such as the following: (1) faster detection of APTs; (2) more accurate detection of APTs, including APTs that are morphing within a family; and/or (3) faster responsiveness to attacks that may be realized by reporting the names of recognized APT and other malware attacks.

More specifically, according to a first embodiment of the disclosure, an electronic device may be implemented with a run-time classifier, which is logic that is capable of accelerating the detection of malware, especially advanced persistent threats (APTs). The run-time classifier is configured to perform, during run-time (e.g., generally contemporaneously with virtual execution operations), an analysis based on (i) anomalous behaviors that are detected during virtual processing of a suspect object within a virtual execution environment and (ii) pre-stored family identifiers. A “family identifier” (also referred to as a “template”) is a collection of data (samples) associated with anomalous behaviors that uniquely identify a particular (APT and/or non-APT) malware family. These anomalous behaviors may constitute (1) unexpected or undesired operations and/or (2) statistically significant usages/accesses of logical components (e.g., files, registry keys, etc.).

The framework for run-time APT analysis comprises one or more databases including family identifiers for APT families and/or malware (non-APT) families. Initially, it is contemplated that the database(s) may be pre-loaded with identifiers associated with currently known APT and non-APT malware families. Thereafter, the database(s) may be updated via one or more external sources and/or in real-time based on results of the APT analysis as described below.

More specifically, as stated above, each family identifier is a collection of data (samples) of anomalous behaviors that uniquely identify a given malware family, namely a collection of related APT malware (referred to as “APT family”) and/or a collection of related malware other than APT malware (referred to as a “non-APT family”. According to one embodiment of the disclosure, the samples of anomalous behaviors may be made generic by removal of the actual arguments (e.g., variable parameters) associated with these behaviors. Hence, this anomalous behavior data (referred to herein as common indicators of compromise “Common IOCs”) may be selected based, at least in part, on the counts maintained for each type of anomalous behavior (IOC) that is associated with the malware forming an entire malware family, namely the related APTs forming a particular APT family or the related malware forming the non-APT family.

For instance, the Common IOCs may be a subset of all samples of anomalous behaviors (IOCs) associated with a particular APT family, where each Common IOC may be generally generated or selected based on (a) removal of actual arguments (parameter values) to make the IOCs generic, and/or (b) filtering out IOCs that would not provide sufficient distinction from other APT families.

The filtering involves removing IOCs (1) with a low occurrence rate with the particular APT family (e.g., less than a first count threshold) and (2) with a high occurrence rate across other known APT families (e.g., greater than a second count threshold). The same technique may be used to generate Common IOCs (family identifiers) for non-APT malware. As a result, Common IOCs are a collection of anomalous behaviors (IOCs) that may be used to uniquely define a given malware family, namely an APT family or a non-APT family.

Stated differently, an APT family identifier for a first APT family, for example, may be generated by obtaining a count of each type of anomalous behavior (IOC) associated with the APTs forming the first APT family, where the count represents the number of occurrences for that anomalous behavior (e.g., IOC). This produces a set of IOCs (e.g., collection of samples of anomalous behaviors) where each IOC may be associated with one or likely more APTs within the first APT family.

Thereafter, the set of IOCs is filtered to remove (i) any IOC from the set of IOCs having a low occurrence rate within the first APT family (e.g., less than the first count threshold) and (ii) any IOC from the set of IOCs having a high occurrence rate across other APT families (e.g., greater than the second count threshold). The later condition ensures entropy among the different APT families to provide sufficient distinctiveness between the APT families. Thereafter, the remaining IOCs, which form a subset of the set of IOCs, are referred to as “Common IOCs” and are used as the “APT family identifier” for the first APT family. A similar process may be conducted to produce a “malware family identifier,” namely a family identifier for a particular (non-APT) malware family.

According to one embodiment of the disclosure, when deployed within an APT detection system, a run-time classifier is configured to initially determine whether anomalous behaviors (IOCs) monitored during virtual processing of a received suspect object within a virtual execution environment statistically matches any (non-APT) malware family identifiers. In other words, the monitored IOCs are compared to the Common IOCs associated with every malware family identifier. Upon detecting a statistical match (e.g. IOCs match 90% or more of the Common IOCs), the analysis for confirming whether the suspect object is an APT is discontinued as the suspect object has now been identified as non-APT malware.

However, if no statistical match is detected, the monitored IOCs are compared with each of the APT family identifiers (e.g. Common IOCs representing each of the APT families). If a statistical match is detected for any of these APT family identifiers (e.g. IOCs match 90% or more of the Common IOCs for a previously classified APT family identifier), the suspect object is considered to be an APT that is part of that previously classified APT family. The family name and/or other stored information associated with the classified APT may be reported to the source submitting the suspect object and/or another electronic device (e.g., network administrator, etc.). This comparison of monitored IOCs with APT family identifiers is performed to provide faster detection of APT malware, as described below.

If no statistical match is detected again, a secondary analysis of the IOCs associated with the suspect object is performed in order to determine whether the suspect object may be classified as some unknown APT (that is not a member of a classified APT family) or malware that is not associated with a classified malware family. This secondary analysis is directed to analyzing the substantive nature of the anomalous behaviors to determine whether these behaviors constitute an APT. For instance, the secondary analysis may review anomalous behaviors involving data theft, statistically significant usages/access of certain logical components such as registry keys), or the like.

After the run-time classifier has completed its analysis, the results may be reported to a targeted destination (e.g., a user of the client device(s), network administrator, etc.) and/or stored in a database. The results may include an identifier for the APT family (hereinafter referred to as the “APT family identifier”), the name of the APT family, monitored behaviors characteristics of the APT family, or the like.

According to a second embodiment, APT family identifiers and/or malware family identifiers may be supplied to an electronic device (e.g., firewall, client device, a threat detection and prevention “TDP” system, etc.) for use in automated detection and prevention of future APT or other malicious attacks. When deployed within the electronic device, a run-time classifier is configured to determine whether anomalous behaviors (IOCs) monitored during virtual processing of a suspect object within a virtual execution environment of the electronic device statistically matches any pre-stored family identifiers such as APT or malware family identifiers. If so, the run-time classifier generates a measurement (referred to as a “score”) and provides the score to a logic unit within the electronic device. The logic unit may use the score, in whole or in part, to determine and signify (to a user, administrator or other entity associated with the source of the suspect object) whether the suspect object is malicious or not. If malicious, based on the finding of the run-time classifier, a name associated with the potential APT or malware family to which the suspect object belongs may be provided.

As an illustrative example, the run-time classifier may be configured to generate a score whose value may be highly correlated to the type of family identifier detected. This score may contribute to the classification of the suspect object as malicious, where the amount of contribution may be based on the weighting applied to this score in determining whether a suspect object is malicious. For instance, the score from the run-time classifier may be aggregated with scores produced from other threat detection processes to produce an overall score that identifies if the suspect object appears to be benign or malware such as APT malware. Alternatively, the score may be utilized in a different manner to potentially influence the overall score.

For instance, when determining that the IOCs suggest that the suspect object is an APT, the run-time classifier may output a first score value. Depending on the weight assigned to scores provided by the run-time classifier (as compared to other scores provided by the other threat detection processes), the first score value may significantly (and perhaps definitely) cause the overall score to represent that the suspect object is malicious. While some embodiments may only use the first score value to signify (perhaps definitively) that the suspect object as malicious, other embodiments may use the first score value to signify (and perhaps definitively) that the suspect object is not only malicious but is an APT.

Also, when determining that the IOCs suggest that the suspect object is not any known malware family, the run-time classifier may output a second score value. Again, depending on the weight assigned, the second score value may have an impact in classifying the suspect object as benign or may have little impact on the classification of the suspect object as malicious.

It is contemplated that the scores output from the run-time classifier may be static for each type of family (e.g. each APT or malware family assigned the same score) or may vary between different types of families (APT, malware) as well as between different types of malware families, between different types of APT families, and between different malware or APTs within their corresponding malware or APT families.

In accordance with another specific implementation, IOCs are stored within a run-time log (e.g., maintained by behavior monitoring logic) of behaviors detected (monitored) during virtual processing of a suspect object within a virtual execution environment and are made generic (prior to storage or thereafter) by removal of actual arguments (parameter values). These monitored behaviors may be used to generate a template (CIOC) as described above. In one embodiment, the logged behaviors may be time-stamped so as to preserve their chronological order during processing and the CIOC is generated to reflect the processing sequence of the CIOC.

II. Terminology

In the following description, certain terminology is used to describe features of the invention. For example, in certain situations, both terms “logic” and “engine” are representative of hardware, firmware and/or software that is configured to perform one or more functions. As hardware, logic (or engine) may include circuitry having data processing or storage functionality. Examples of such circuitry may include, but is not limited or restricted to a microprocessor, one or more processor cores, a programmable gate array, a microcontroller, an application specific integrated circuit, wireless receiver, transmitter and/or transceiver circuitry, semiconductor memory, or combinatorial logic.

Logic (or engine) may be software in the form of one or more software modules, such as executable code in the form of an executable application, an application programming interface (API), a subroutine, a function, a procedure, an applet, a servlet, a routine, source code, object code, a shared library/dynamic load library, or one or more instructions. These software modules may be stored in any type of a suitable non-transitory storage medium, or transitory storage medium (e.g., electrical, optical, acoustical or other form of propagated signals such as carrier waves, infrared signals, or digital signals). Examples of non-transitory storage medium may include, but are not limited or restricted to a programmable circuit; a semiconductor memory; non-persistent storage such as volatile memory (e.g., any type of random access memory “RAM”); persistent storage such as non-volatile memory (e.g., read-only memory “ROM”, power-backed RAM, flash memory, phase-change memory, etc.), a solid-state drive, hard disk drive, an optical disc drive, or a portable memory device. As firmware, the executable code is stored in persistent storage.

The term “object” generally refers to a collection of data, whether in transit (e.g., over a network) or at rest (e.g., stored), often having a logical structure or organization that enables it to be classified for purposes of analysis. The objects may be associated with network traffic. During analysis, for example, the object may exhibit a set of expected characteristics and, during processing, a set of expected behaviors. The object may also exhibit a set of unexpected characteristics and a set of unexpected behaviors that may evidence malware and allow the object to be classified as at least “malicious” and perhaps classified as an advanced persistent threat (APT), when warranted.

Examples of objects may include one or more flows or a self-contained element within a flow itself. A “flow” generally refers to related packets that are received, transmitted, or exchanged within a communication session. For convenience, a packet broadly refers to a series of bits or bytes having a prescribed format, which may include packets, frames, or cells. A “message” may be broadly referred to as any series of bits or bytes having a prescribed format as well.

As an illustrative example, an object may include a set of flows such as (1) a sequence of transmissions in accordance with a particular communication protocol (e.g., User Datagram Protocol (UDP); Transmission Control Protocol (TCP); or Hypertext Transfer Protocol (HTTP); etc.), or (2) inter-process communications (e.g., Remote Procedure Call “RPC” or analogous processes, etc.). Similar, as another illustrative example, the object may be a self-contained element, where different types of such objects may include an executable file, non-executable file (such as a document or a dynamically link library), a Portable Document Format (PDF) file, a JavaScript file, Zip file, a Flash file, a document (for example, a Microsoft Office® document), an electronic mail (email), downloaded web page, an instant messaging element in accordance with Session Initiation Protocol (SIP) or another messaging protocol, or the like.

As noted above, an APT is a type of sophisticated network attack that is directed at a particular target and seeks to surveil, extract, and/or manipulate data to which a targeted entity would have access. APTs may seek to maintain a persistent attack on a targeted electronic device and may initially lay dormant (prior to activation) for a prolonged period of time in comparison with traditional malware.

For example, a self-contained element of a flow, such as an APT-latent email message for example, may be specifically directed to a particular individual at a company (e.g., an officer of the company) in an attempt to extract sensitive data accessible by that individual. Sometimes, the APT-latent email message may include text/greetings that are personalized for the targeted entity along with an attachment (e.g., a Portable Document Format (PDF) document). The attachment may contain malicious content such that, upon opening or otherwise activating the attachment, the malicious content attempts to extract and/or manipulate targeted data accessible to the defined target.

Malware may be construed broadly as software that, upon execution, is designed to take advantage of a vulnerability, for example, to harm or co-opt operation of an electronic device or misappropriate, modify or delete data as for APT malware. Conventionally, malware is often said to be designed with malicious intent. An object may constitute or contain malware, whether APT malware or non-APT malware.

The term “transmission medium” is a physical or logical communication path between two or more electronic devices (e.g., any devices with data processing and network connectivity such as, for example, a security appliance, a server, a mainframe, a computer such as a desktop or laptop, netbook, tablet, firewall, smart phone, router, switch, bridge, etc.). For instance, the communication path may include wired and/or wireless segments. Examples of wired and/or wireless segments include electrical wiring, optical fiber, cable, bus trace, or a wireless channel using infrared, radio frequency (RF), or any other wired/wireless signaling mechanism.

In general, a “virtual machine” (VM) is a simulation of an electronic device (abstract or real) that is usually different from the electronic device conducting the simulation. A VM may be used to provide a sandbox or safe runtime environment to enable detection of APTs and/or other types of malware in a safe environment. The VM may be based on specifications of a hypothetical computer or emulate the computer architecture and functions of a real world computer.

The term “computerized” generally represents that any corresponding operations are conducted by hardware in combination with software and/or firmware. Also, the term “compare” or “comparison” generally means determining if a statistical match (e.g., a certain level of correlation) is achieved between two items where one of the items may include a particular family identifier, as described below in detail.

Lastly, the terms “or” and “and/or” as used herein are to be interpreted as inclusive or meaning any one or any combination. Therefore, “A, B or C” or “A, B and/or C” mean “any of the following: A; B; C; A and B; A and C; B and C; A, B and C.” The phrase “(A, B, . . . , etc.)” has a similar connotation. An exception to this definition will occur only when a combination of elements, functions, steps or acts are in some way inherently mutually exclusive.

The invention may be utilized for detection, verification and/or prioritization of malicious content such as exploits. As this invention is susceptible to embodiments of many different forms, it is intended that the present disclosure is to be considered as an example of the principles of the invention and not intended to limit the invention to the specific embodiments shown and described.

III. General Operation Flow of Run-Time Malware Classification

Referring to FIG. 1, an exemplary block diagram of an operational flow of a run-time malware classification conducted within an electronic device 100 is shown. Herein, incoming objects 120 originally associated with network traffic are uploaded into a virtual environment 130. Herein, the virtual environment 130 comprises virtual execution logic including one or more virtual machines that virtually process (also referred to as “detonate”) each of the incoming objects 120. The virtual environment 130 further monitors behaviors during such virtual processing. Some or all of these monitored behaviors 140 are provided to a run-time classifier 150 for analysis in real-time. For this embodiment, only anomalous behaviors, namely unexpected or undesired operations by the suspect object and/or statistically significant usages/access of certain logical components (e.g., registry keys, certain ports or files, etc.), are provided to the run-time classifier 150. Of course, all monitored behaviors may be provided to the run-time classifier 150, which would be provided functionality for identifying the anomalous behaviors from normal behaviors.

According to one embodiment of the disclosure, the run-time classifier 150 is configured to initially determine whether the anomalous behaviors 140 (sometimes referred to as “indicators of compromise” or “IOCs), being part of the monitored behaviors during virtual processing of an object, statistically match one of a plurality of preconfigured family identifiers stored in database 160. For this embodiment, a family identifier may be either (i) an APT family identifier directed to a particular APT family or (ii) a malware family identifier directed to a non-APT malware family.

Herein, according to one embodiment of the disclosure, the family identifier database 160 may follow a relational, object, hierarchical, or any other type of database model. In one embodiment, the family identifier database 160 is spread across one or more persistent data stores. The persistent data stores may be integrated within the electronic device 100 (e.g., APT detection system 200 or TDP system 910₁ described below) or within a separate host electronic device. For example, the family identifier database 160 may be located at a remote or even geographically remote location that is communicatively coupled (e.g., by a dedicated communication link or a network) to the electronic device 100.

As stated above, each family identifier is a collection of samples of anomalous behaviors, also referred to herein as common indicators of compromise (“Common IOCs”). The Common IOCs may be selected based, at least in part, on the counts maintained for each type of anomalous behavior (IOC) that is associated with the APTs (or malware) forming a particular family (e.g., APT family, malware family, etc.).

Therefore, if the IOCs associated with the suspect object statistically match any Common IOCs corresponding to the family identifiers, the run-time classifier 150 determines that the suspect object is part of that particular (APT or malware) family. Depending on the deployment for the run-time classifier, a number of actions may be undertaken by the electronic device when the IOCs statistically match any Common IOCs representing a family identifier. For instance, as an example, the particulars associated with the uncovered family may be reported, as represented by output 170. As another example, further analysis in determining whether the suspect object is an APT may be discontinued or may be continued to obtain further analytical information. As yet another example, a score associated with the uncovered family may be provided to logic within the electronic device that is responsible for deciding whether the incoming suspect object is malicious or not, as represented by output 180. The determination logic can rely solely on the determination or weigh other considerations when making the decision.

If no statistical match is detected, a secondary analysis of the IOCs associated with the suspect object may be performed to determine whether the suspect object still may be classified as an APT (e.g., the suspect object is an APT that does not belong to any classified APT or malware families, as represented by output 190.

Hence, the run-time classifier 150 is useful to more quickly detect APTs and other types of non-APT malware. This may enable network administrators to address malicious attacks before serious security issues may arise.

IV. APT Detection System Deployment of Run-Time Classifier

Referring to FIG. 2, an exemplary block diagram of a first illustrative embodiment of an APT detection system 200 that is adapted to perform run-time APT classification on incoming objects 220 is shown. Received from a source via network 210, the incoming objects 220 are directed to virtual execution logic 230, which is part of the APT analysis system 330 as shown in FIG. 3.

Herein, the incoming objects 220 may be captured and filtered from some or all of the incoming objects associated with network traffic. For instance, as an illustrative example, before receipt of an incoming object (e.g., one of incoming objects 220) by virtual execution logic 230, it is contemplated that intrusion protection system (IPS) logic or heuristic logic (not shown) may be deployed to filter incoming objects 220 from a larger set of objects (not shown). Such filtering may be conducted through signature checks or other static analysis processes, where the incoming objects 220 are tagged for more in-depth analysis. Given that the source (not shown) may be an electronic device that has already determined that some or all of the incoming objects 220 may be malicious, the operations of the IPS logic and/or heuristic logic may be suspended or the amount of filtering realized may be de minimis. In fact, for some deployments, the IPS/heuristic logic is unnecessary and may not be implemented.

Herein, the virtual execution logic 230 comprises one or more virtual machines 235₁-235_N (N≧1), which virtually processes (sometimes referred to as “detonate”) each of the incoming objects 220 and monitors for anomalous behaviors during such virtual processing, as described below. These anomalous behaviors 240 are provided to the run-time classifier 150 for analysis.

According to one embodiment of the disclosure, the run-time classifier 150 is configured to initially determine whether the anomalous behaviors 240 (IOCs) statistically match any malware family identifiers. According to this embodiment of the disclosure, the malware family identifiers may be retrieved from malware family identifier database 162, which along with an APT family identifier database 164 forms the family identifier database 160. Family identifier database 160 may be located within the APT detection system 200 or may be located at remote or even geographically remote location that is communicatively coupled (e.g., by a dedicated communication link or via network 210) to the APT detection system 200 (e.g., cloud services; dedicated external server; etc.). Both the malware family identifiers and the APT family identifiers may be updated in a periodic or aperiodic manner, based on data downloaded from an external server (not shown) or data from suspect objects 220 detected as APTs or other malware by run-time classifier 150.

As stated above, each family identifier is a collection of data (samples) associated with anomalous behaviors that uniquely identify a given APT (or malware) family. This collection of anomalous behavior data (Common IOCs) may be selected based, at least in part, on the counts maintained for each type of anomalous behavior that is associated with the APTs (or malware) forming a particular APT (or malware) family. For instance, the Common IOCs (S1, S2, S4, S8 and S18) representing a first APT family identifier for a first APT family may be a subset of samples S1-S20 of anomalous behaviors for the first APT family. Each Common IOC (S1, S2, S4, S8 and S18) may be selected based on (1) a high occurrence rate of an IOC (e.g., greater than a first count threshold) for the first APT family and (2) a lower occurrence rate of this behavior (e.g., less than a second count threshold) across other APT families stored in APT family identifier database 164. Similarly, the Common IOCs (S3-S4, S10 and S28) representing a second APT family identifier may be a subset of samples (S1-S5, S10-S12 and S21-S30) of anomalous behaviors for the second APT family. It is noted that some Common IOCs may be shared between different APT family identifiers, provided that the Common IOCs in their entirety are collectively distinct and unique.

Therefore, if the IOCs associated with the suspect object statistically match any of the malware family identifiers retrieved from malware family identifier database 162, the run-time classifier 150 determines that the suspect object is not an APT and discontinues its analysis.

If no statistical match is detected, the monitored IOCs from the suspect object are compared with each of the APT family identifiers retrieved from APT family identifier database 164. If a statistical match is detected, the suspect object is considered to be an APT that is part of a previously classified APT family. Information 250 associated with the classified APT family (e.g., family name, suspect object, Common IOCs used, etc.) may be received by reporting logic 260 and forwarded to the source submitting the object or to another electronic device (e.g., administrator, etc.), as denoted by flow 265.

If no statistical match is detected, a secondary analysis of the IOCs associated with the suspect object may be performed by a secondary classifier 280 that receives at least the anomalous behaviors 270 to determine whether the suspect object may be classified as an APT or not. This secondary analysis may involve substantive review for anomalous behaviors directed to data theft, statistically significant usages/access of certain logical components such as registry keys, or the like. Hence, the run-time APT analysis is used prior to this analysis to achieve faster detection of APTs associated with APT families that have already been classified, as further described below.

Referring now to FIG. 3, an exemplary block diagram of an illustrative embodiment of a communication system 300 is shown, where a source 310 may upload suspect objects to the APT detection system 200 for analysis as to whether each of these suspect objects is an APT. Herein, the communication system 300 comprises APT detection system 200 communicatively coupled to the source 310 over transmission medium forming the network 210. In general, according to one embodiment of the disclosure, the APT detection system 200 comprises one or more electronic devices that are configured to receive one or more suspect objects 320 from the source 310 (e.g., client devices 310A and 310B) for APT detection and potential APT family classification.

More specifically, according to this embodiment, the APT detection system 200 comprises an APT analysis system 330, an APT server 360, and the family identifier database 160. In particular, the APT server 360 operates in combination with the family identifier database 160 and/or APT analysis system 330 to automatically determine whether an incoming suspect object 320 is an APT belonging to a previously classified APT family.

According to one embodiment of the disclosure, the suspect object 320 is provided to the APT analysis system 330, in particular the virtual execution logic 230 within the APT analysis system 330. The virtual execution logic 230 comprises a run-time virtual execution environment 340 that comprises one or more virtual machines (VMs) 345₁-345_M (M≧1), where one or more of the VMs 345₁-345_M may be configured for virtual processing the suspect object 320 which may cause anomalous behaviors to occur.

Although not shown, VMs 345₁-345_M may be configured based on the results of the signature checks conducted prior to routing the subject object 320 to the APT analysis system 330. Alternatively, metadata associated with the subject object 320 may be used, at least in part, to determine protocols, application types and other information that may be used to determine particular software profile(s). The software profile(s) are used for selecting corresponding software images within a data store 335 for use in configuring a run-time environment in the one or more virtual machines 345₁-345_M. These software profile(s) may be directed to different versions of the same software application for fetching corresponding software image(s) from data store 370.

During virtual execution of the subject object by one or more of the VMs 345₁-345_M, the behaviors exhibited during virtual processing are monitored by a behavior monitoring logic 350. Of these monitored behaviors, a count may be maintained by behavior counter 365 (deployed in APT analysis system 330 or APT server 360 as shown) for at least each type of monitored anomalous behavior 352. The anomalous behaviors 352 are provided from APT analysis system 330 to the run-time classifier 150 of the APT server 360. The dropped object extractor 355 performs operations to detect, extract, and pass dropped objects during virtual processing by the suspect object 320 by VM(s) 345₁, . . . and/or 345_M.

As illustrated in FIG. 3, the APT server 360 comprises the behavior counter 365, an APT classifier 370, a warning generator 380, and reporting logic 260. Herein the APT classifier 370 includes the run-time classifier 150, secondary classifier 280 and family identifier generator 375. While run-time classifier 150 and secondary classifier 280 are configured to attempt to classify suspect objects 320 as APTs based on an analysis relying on family identifiers, the family identifier generator 375 is configured to generate such family identifiers.

Referring to FIG. 4, in accordance with formulation of the framework for conducting the run-time APT analysis using the family databases, an illustrative embodiment of operations conducted by the family identifier generator 375 to generate a family identifier, such as an APT family identifier for example, is shown. Initially, the family filter generator obtains samples of all types of anomalous behaviors (IOCs) associated with the APTs forming a particular APT family along with their corresponding counts (operation 400). These IOCs may be referred to as a set of IOCs. Upon obtaining the set of IOCs, the family filter generator performs a first filtering operation by eliminating any IOC that falls below a first occurrence rate within the particular APT family to produce a first subset of IOCs (operation 410). The first occurrence rate may represent a first count threshold, which may be a static value or a dynamic value.

Thereafter, the family filter generator performs a second filtering operation on the first subset of IOCs by eliminating any IOC having a second occurrence rate within APT families other than the particular APT family (operation 420). The second filter operation produces a second subset of IOCs. Herein, the second occurrence rate may represent a second count threshold, which may be greater (and perhaps substantially greater by a few factors) than the first count threshold. Similarly, the second count threshold may be a static value or a dynamic value. Of course, it is contemplated that the second filtering operation may be conducted prior to the first filtering operation as the ordering of these operations may be changed providing both filtering operations are performed.

After performing the first and second filtering operations, the second subset of IOCS may constitute the Common IOCs that represent the APT family identifier for the particular APT family (operation 430). Of course, based on the number of IOCs forming the second subset of IOCs, it is contemplated that only some of the second subset of IOCs may be used as the APT family identifier.

Referring back to FIG. 3, after receipt of the anomalous behaviors 352 (e.g., IOCs) associated with the suspect object 320 from APT analysis system 330, the run-time classifier 150 within the APT server 360 determines if the suspect object corresponds to any malware family identifiers (e.g. corresponds to a predetermined percentage of CIOCs forming the malware family identifier), which may be obtained from the malware family identifier database 162. In general, this determination involves a statistical comparison of the IOCs associated with the suspect object 320 to the malware family identifiers within the malware family identifier database 162. Upon determining that the IOCs associated with the suspect object 320 statistically match any of the malware family identifiers, the run-time classifier 150 discontinues analysis on the suspect object 320 as it has been classified other than an APT. For instance, a “statistical match” may be a determination that ninety percent (90%) or more of the IOCs match the compared Common IOCs as described above.

Upon failing to detect a statistical match between the IOCs associated with the subject object 320 and the malware family identifiers, the run-time classifier 150 analyzes these IOCs in connection with the APT family identifiers, which may be retrieved from APT family identifier database 164. Upon comparing the IOCs with some or all of APT family identifiers and detecting a statistical match, the run-time classifier 150 has identified the suspect object 320 as an APT that is part of the classified APT family. It is contemplated that, for testing purposes, the suspect object 320 may be an APT from a known APT family in order to better define APT family boundaries.

In response to detecting that object 320 is an APT of a classified APT family, the warning generator 380 of the APT server 360 generates and transmits a warning message 385 to the source 310 (e.g., a corresponding client device 310A). The warning message 385 may indicate to a targeted recipient (e.g., client, IT personnel, etc.) that the suspect object 320 is an APT, perhaps along with its determined APT family name; the APT family identifier for use in detecting future attacks, namely the Common IOCs representing the APT family (e.g., anomalous behaviors such as data theft, statistically significant usages/access of certain logical components such as registry keys); or the like. Alternatively, the warning message 385 may be routed to another electronic device (e.g., administrator, etc.).

If no statistical match is still detected by run-time classifier 150, a secondary classifier 280 is now provided with the IOCs associated with the subject object 320 and analyzes the substantive nature of these IOCs to determine whether the suspect object may be classified as an APT or not.

Referring still to FIG. 3, one or more client devices 310A and 310B are coupled to the APT detection system 200 through the network 210. Network 210 may be a private network (e.g., enterprise network) in which both the APT detection system 110 and the client devices 310A and 310B are on the same network. Alternatively, network 210 may be a public network in which the APT detection system 200 is remotely accessed by an electronic device (e.g., client 310A/310B, etc.).

Herein, the client device(s) 310A/310B may be any type of electronic device, including laptop computers, desktop computers, tablet computers, smartphones, servers, network devices (e.g., firewalls and routers), wearable technology, process controllers, or other types of electronics with data processing capabilities and typically have network connectivity. Furthermore, the client device(s) 310A/310B may include one or more processors with corresponding memory units for processing data. The processors and memory units are generally used herein to refer to any suitable combination of programmable data processing components and data storage that conduct the operations needed to implement the various functions and operations of the client device(s) 120. The processors may be special purpose processors such as an application-specific integrated circuit (ASIC), a general purpose microprocessor, a field-programmable gate array (FPGA), a digital signal controller, or a set of hardware logic structures (e.g., filters, arithmetic logic units, and dedicated state machines) while the memory units may refer to non-volatile memory. An operating system may be stored in the memory units of the client device(s) 310A/310B, along with application programs specific to the various functions of the client device(s) 310A/310B, which are to be run or executed by the processors to perform the various functions of the client device(s) 310A/310B. For example, the memory units of a client device 310A/310B may store email and/or web-browser applications that are run by associated processors to send, receive, and view information associated with the objects.

FIG. 5 shows a component diagram of the APT server 360 according to one embodiment of the invention. As shown, the APT server 360 may include one or more processors 500 and a persistent data store 530, where processor(s) 500 is further coupled to persistent storage 530 via transmission medium 525.

The one or more processors 500 and the persistent data store 530 are generally used herein to refer to any suitable combination of programmable data processing components and data storage that conduct the operations needed to implement the various functions and operations of the APT server 360. The processor(s) 500 may be one or more special purpose processors such as an application-specific integrated circuit (ASIC), a general purpose microprocessor, a field-programmable gate array (FPGA), a digital signal controller, or a set of hardware logic structures (e.g., filters, arithmetic logic units, and dedicated state machines) while the persistent data store 530 may refer to non-volatile memory. An operating system may be stored in the persistent data store 530, along with application programs specific to the run-time classifier 150 and other various functions of the APT server 360, which are to be run or executed by the processors 500 to perform the various functions of the APT server 360.

In one embodiment, the APT server 360 may include one or more input/output (I/O) interfaces 510 for communicating with various components external to the APT server 360. The I/O interface(s) 510 may be implemented as a physical interface including one or more ports for wired connectors. Additionally, or in the alternative, I/O interface 510 may be implemented with one or more radio units for supporting wireless communications with other electronic devices. Hence, the I/O interface(s) 510 enables communications with one or more electronic devices using wireless and/or wired protocols, including the IEEE 802.3 and the IEEE 802.11 suite of standards.

In one embodiment, as shown in FIG. 5, the I/O interface(s) 510 allows the APT server 360 to communicate with the family identifier database 160, an external server 540, APT analysis system 330, and/or the source 310 for suspect objects over one or more wired and/or wireless transmission mediums. It is contemplated that the APT analysis system 330 may be optional where the virtual processing of the suspect object occurs outside the APT detection system 200, and anomalous behaviors are provided to APT server 360 from other resources (including remote sources.

Referring still to FIG. 5, the persistent data store 530 may store logic, including the run-time classifier 150, the secondary classifier 280, the warning generator 380 and reporting logic 260 (e.g., a graphical user interface “GUI”). Each of these logic elements may be discrete software components that may be processed/run by one or more of the processors 500. Each element stored in the persistent data store 530 and shown in FIG. 5 will be described below in accordance with the method of operation described in FIGS. 6A-6B.

Referring to both FIGS. 5-6A, an illustrative embodiment of a method for identifying and classifying APT objects is shown. Herein, at operation 500, a suspect object is received by the APT detection system 200, namely the virtual execution logic within the APT analysis system 330 receives the suspect object 320 from the source 310 (e.g., client device 310A, a threat detection and prevention “TDP” system of FIG. 9, etc.). It is contemplated that, where the source 310 is a TDP system which also analyzes the suspect object for malware, the suspect object is provided directly to the APT detection system 200 to verify whether the suspect object includes APT malware. However, where the source 310 is the client device 310A, it may utilize the APT detection system 200 as a primary malware detection system, and thus, APT server 360 may include graphics user interface (GUI) logic 550 that allow a user of the client device 310A to submit a suspect object through an interface. The interface may be generated by the GUI logic 550 and served to the client device 310A. In this fashion, the APT server 330 may operate as a web-server to deliver data and provide a user interface to the client device 310A.

Referring to FIG. 7A, an exemplary web-interface 700 for submitting a suspected object to the APT server 360 from client device 310A is shown. In this example interface, a user may direct a web browser running on the client device 310A to view the web-interface 700. The user may thereinafter enter the address/location of a suspect object into the web-interface 700 using the address input field 710 and the “BROWSE” button 720. The entered address indicates the location of the suspect object in storage on the client device 310A or on a remote device (e.g., stored on a server). After selection of the suspect object, the user may submit the suspect object to the APT server 360 by selecting the “SCAN” button 730 in the web-interface 700. The suspect object may be transmitted from the client device 310A such that it is received by the APT server 360 for processing as described above at operation 600.

Although the APT server 360 is described above to serve the web-interface 700 to a browser of the client device 310A, in other embodiments, a separate web-server may be in communication with the client device 310A and the APT server 360 to provide the web-interface 700 and facilitate transmission of the suspect object to the APT server 360 from the client device 310A.

Referring back to FIGS. 5-6A, following receipt of the suspect object, APT analysis system 330 detonates the suspect object (e.g., processes by virtual execution or other operations to activate the suspect object) to produce data describing the anomalous behaviors of the suspect object during virtual processing (operation 605). In one embodiment, the APT analysis system 330 comprises one or more separate computing devices or processing units that may independently and discretely process the suspect object and monitor (e.g., log, count, etc.) the resultant operations.

For example, in one embodiment, the suspect object may be a self-contained element of a flow such as a PDF file. In this embodiment, APT analysis system 330 may configure a VM with Adobe® Reader® or other appropriate document reader to detonate the PDF file (e.g., performs virtual operations on the PDF file). The resultant behaviors performed during virtual processing of the suspect object are monitored by behavior monitoring logic 375, and a count (number of occurrences) for each type of monitored, anomalous behavior may be maintained. Each of these counts may include the number of occurrences of a particular anomalous behavior during virtual processing of the suspect object and/or associated dropped objects (hereinafter referred to as “behavior count value”).

After detonating the suspect object, the APT analysis system 330 monitors and records at least the anomalous behaviors and their aggregated behavior count values. This recorded data describing the suspect object. Use of the APT analysis system 330 ensures that detonation of the suspect object is controlled and will not result in infection of the client device 310A and/or the compromise of sensitive data. According to one embodiment, the APT analysis system 330 conducts heuristic analysis.

According to another embodiment, the APT analysis system 330 includes a plurality of VMs with various profiles, and may, in some cases, simulate the client device 310A during detonation of the suspect object. These profiles may include software to be run by a VM to process a suspect object. For example, the profiles may include an operating system and one or more suitable computer applications that are required to process the objects. In this example, one of the applications may include a document reader (e.g., an Adobe® Reader for PDF documents) and/or a web browser (for web pages) for detonating the suspect object. The APT analysis system 330 may include separate virtual processors and memory units for use in detonating different types of objects.

As noted above, detonation of the suspect object at operation 305 produces data that describes anomalous behaviors of the suspect object. Of course, besides data directed to the anomalous behaviors and their corresponding count values, the data may also include details regarding the origin of the suspect object stored in metadata, data generated by the suspect object during detonation, data attempted to be accessed by the suspect object (both locally and from remote systems) during detonation, etc.

During detonation, in some cases, the suspect object may generate/drop separate objects during detonation. These dropped objects may be new files (e.g., binary files) or other segments of data or executable code created by the original suspect object. In this embodiment, the dropped objects may be uncovered and passed back to operation 605 for detonation by the dropped object extractor (operations 610 and 615). Accordingly, each of the dropped objects is detonated in a similar fashion as described in relation to the suspect object to generate data associated with each dropped object and behavior count values for behaviors detected during analysis of the original suspect object may be augmented to reflect the actual number of occurrences for each particular behavior.

As shown in operation 620, after detonation of the suspect object and any dropped objects produced by the suspect object, anomalous behaviors associated with the suspect and dropped objects may be determined from the data. Additionally, the behavior count values may be tallied or, in the alternative, determined by the run-time classifier 150 in lieu of the APT analysis system 330 performing the behavior count analysis.

As an illustrative example, monitored anomalous behaviors of the objects during detonation along with the number of occurrences of these particular behaviors (behavior count value) are maintained. For instance, a first behavior count value associated with a first anomalous behavior may represent the number of occurrences that a suspect object attempts to make out-bound communications to outside data sources during virtual processing of that object. Outbound communications may seek instructions, for example from a malicious server, as to conduct malicious activity. In another embodiment, a second behavior count value associated with a second anomalous behavior may represent the number of occurrences that a suspect object is attempting to exfiltrate (or send out) data to an external resource. In fact, exfiltration of data alone may indicate that the object is an APT.

Hence, the anomalous behaviors provide a comprehensive description of an associated object such that a comparison of anomalous behaviors, and analysis of their corresponding count values may be performed. Such comparison/analysis is conducted to determine whether the object is an APT and/or belongs to a particular APT family, as described below.

Optionally, data related to the suspect object and the dropped objects may be further retrieved from external data sources while anomalous behaviors are being monitored during virtual processing of the suspect object. For example, data may be retrieved from the external server 540 through the I/O interface 510. In this embodiment, the external server 540 may be a device on the same local area network as the APT detection system 200 or connected to the APT detection system 200 over a wide area network (e.g., the Internet). For example, as discussed above, the external server 540 may be connected to the APT detection system 200 through the network 210 of FIG. 3.

In one embodiment, the data retrieved from the external server 540 may include data related to servers attempted to be accessed by the suspect and dropped objects while being detonated (e.g., internet protocol (IP) address of a server) where APT analysis system 200 physically processes the object in lieu of logical (virtual) processing. In another embodiment, the external data may include data collected by third parties related to the suspect object (e.g., malware classification information). In one embodiment, operation 620 may be performed by the run-time classifier 150.

Following generation of behaviors for the suspect object and/or the dropped objects, as shown in operation 625, the APT classifier 250 may analyze the data to automatically (1) determine whether the suspect object belongs to a known malware (non-APT) family. If not, the APT classifier 250 may determine (i) whether the suspect object is an APT belonging to a previously classified APT family, or (2) whether the suspect object is an APT where a family associated with the suspect object has not been classified.

More specifically, as shown in FIG. 6A, the run-time classifier 150 performs an analysis of the anomalous behaviors (IOCs) associated with the suspect object to any or all of the malware family identifiers (block 525). This analysis may involve statistical comparison of the IOCs associated with the suspect object to the Common IOCs formulating each of the malware family identifiers. If a statistical match is detected, the run-time classifier discontinues further processing of the suspect object as the object is now determined to be malware other than an APT (operations 630 and 635). However, it is contemplated that the analytic data may be generated and reported. Otherwise, if no statistical match is detected, the run-time classifier continues further analysis (operation 640).

Referring now to FIG. 6B, after conducting a first analysis by statistically comparing the IOCs of the suspect object to the Common IOCs associated with the malware family identifiers, the run-time classifier conducts a second analysis by comparing the IOCs of the suspect object to the Common IOCs associated with previously classified APT families as retrieved from the family identifier database (operation 645). If a statistical match is detected, the object is an APT and is classified as belonging to that APT family (blocks 650 and 655).

In one embodiment, each entry in the APT family identifier database 164 may include the suspect object along with the APT family identifier that uniquely identifies the object in the APT family identifier database. Other information that may be stored within APT family identifier database 164 may include one or more of the monitored anomalous behaviors (e.g., IOCs) for the suspect object, data from external server or other resources, or the like (operation 655).

Upon determining that the suspect object is APT malware and storage of its behaviors (IOCs), the suspect object is flagged as APT malware in the APT family identifier database (operation 660). In one embodiment, flagging the suspect object as APT malware includes setting an APT malware data value associated with the suspect object in the APT family identifier database 164 to a selected value, e.g., “true”. Also, the APT family identifier is stored to identify that the APT malware belongs to a certain APT family from which information associated with the APT family can be accessed for inclusion within the warning message or subsequently accessed by the user or administrator using the APT family identifier.

After flagging the suspect object as APT malware in the APT family identifier database, the warning generator within the APT server generates a warning message to a targeted destination such as a source of the suspect object (e.g., client device 310A or TDP system) or another electronic device (operation 665). The warning message identifies that the suspect object is APT malware and should be discarded, deleted, quarantined or otherwise avoided.

By the warning generator 389 transmitting a warning message or other messages to the source identifying a classification of the suspect object, a user or administrator of the source may be better prepared and less susceptible to APTs and other malicious attacks. For example, upon receiving a warning message from the APT detection system 200 of FIG. 3, the user may delete/quarantine the suspect object(s) (e.g., an email or file) and/or report the suspect object(s) to a network administrator. Also, the APT detection system 200 may generate the APT family identifier for the APT malware including its metadata, such as, for example, its behaviors observed during processing. The APT family identifiers may be stored in the APT family identifier database 162 and may be distributed to one or more other electronic devices. The APT family identifier (or parts thereof) may be used to generate a signature for the APT malware, which may be used in turn by the source or other electronic devices to block future objects/content where signature statistically matches are found. This proactive action may prevent the source from being infected by the suspect object and sensitive data accessible to the user is not compromised by the suspect object.

In one embodiment, the warning message may be a transmission to a component of the web-interface 700. For example, as shown in FIG. 7B, a dialog box 750 of the web-interface 700 may be updated to indicate that the suspect object is APT malware. In other embodiments, other warnings may be transmitted to the client device 310A. For example, email messages, pop-up messages, or other signals may be transmitted between the APT detection system 200 and the client device 310A to represent the warning message.

Similarly, upon determining at operation 650 that the suspect object is not APT malware, the run-time classifier stores the suspect object, some or all of the anomalous behaviors (IOCs) associated with the suspect object, and/or data from external sources into the APT family identifier database 164 (operation 670). Thereafter, the secondary classifier is configured to determine whether the suspect object is APT malware or non-APT malware based on comparisons with anomalous behaviors of the suspect object to highly common anomalous behaviors (operation 675). This comparison may be performed using machine learning and statistical analysis.

Upon determining that the suspect object is non-APT malware, the secondary classifier 280 flags the suspect object as non-APT malware (operation 685). In one embodiment, flagging the suspect object as non-APT malware includes setting an APT malware data value associated with the suspect object temporarily stored the APT family identifier database 164 to a selected value, e.g., “false”. However, upon determining that the suspect object is new APT malware, the suspect object is flagged as APT malware in the APT family identifier database 164 (operation 685), where the APT may be assigned to a new APT family identifier or assigned to a “known” classification for now. Thereafter, the analytic data has been generated (operation 690).

V. Threat Detection and Protection System (TDP) System Deployment of Run-Time Classifier

Although described above as transmission of a suspect object, in other embodiments, a suspect object may be analyzed separate from the APT detection system 200, where the monitored APT detection functionality deployed within an electronic device (e.g., firewall, client device, a threat detection and prevention “TDP” system, etc.). According to a second embodiment of the disclosure, as shown in FIG. 8, the electronic device 800 may be adapted to receive the APT family identifiers and/or malware family identifiers for use in automated detection and prevention of future APT or other malicious attacks at the appliance level.

In contrast to deployment within the APT detection system, when deployed within the electronic device 800, a run-time classifier 850 may be configured to determine whether anomalous behaviors (IOCs) monitored during virtual processing of a suspect object within a virtual execution environment statistically matches any pre-stored APT or malware family identifiers within family identifier database 160. If so, the run-time classifier 850 generates a measurement (referred to as a “score”) that is provided to the score determination logic 860 within the electronic device. The score determination logic 860 may use the score, in whole or in part, in determining whether the suspect object is to be classified malicious or not.

As an illustrative example, received from a source via network 210, incoming objects 805 are captured and subsequently analyzed by static analysis logic 810 to (i) filter a subset of the objects 820 from the incoming objects 805 and/or (ii) generate a score (Score_1) 815 associated with each object that reflects the likelihood of the object being malware (and perhaps the severity of the potential malware).

In particular, as an illustrative example, before receipt of objects 820 by virtual execution logic 825, the static analysis logic 810 (e.g., IPS logic, heuristic logic) may conduct signature checks (e.g., exploit signature checks, vulnerability signature checks, etc.) or other scanning operations on the objects 805, where a subset of objects 820 are tagged for more in-depth analysis. Furthermore, the static analysis logic 810 may be configured to generate a score (Score_1) 815 for each analyzed object, which represents the probability (or level of confidence) that the characteristics of that analyzed object are indicative of malware. In other words, the score represents a value that classifies the threat level of the possible malware characterized by the particular analyzed object.

For instance, as an illustrative example, upon detecting one type of characteristic that suggests an object 820 under analysis is malware, the static analysis logic 810 may generate a score having a first value (e.g., score of 5 out of 20) associated with that object. However, upon detecting multiple characteristics or another type of characteristic that more strongly suggests the object under analysis is malware, a higher score (e.g., score of 13 out of 20) may be generated.

Herein, the virtual execution logic 825 comprises a run-time virtual execution environment 830 that features one or more virtual machines 835₁-835_N (N≧1), which virtually processes (sometimes referred to as “detonate”) each of the incoming objects 820. Behavior monitoring logic 840 monitors the behaviors produced during virtual processing of a suspect object 820 and determines which the these behaviors are anomalous. These anomalous behaviors 845 are provided to the run-time classifier 850 for analysis.

The run-time classifier 850 may be configured to generate a score (Score_2) 855 whose value may be dependent on whether the suspect object is classified to be part of a known malware (non-APT or APT) family and/or the type of malware family. Score_2 855 may contribute to the classification of the suspect object as malicious, where the amount of contribution may be based on the weighting applied to Score_2 855. For instance, Score_2 855 may be aggregated with scores produced from other threat detection processes (e.g., Score_1 produced by static analysis logic 810) or may be utilized in a different manner to influence the overall score used to identify whether the suspect object is malicious or not. The score determination logic 860 generates the overall score 865 to an object classifier 870 that identifies to reporting logic 880 within electronic device 800 if the suspect object appears to be benign, non-APT malware or APT malware.

For instance, when determining that the anomalous behaviors (IOCs) 845 suggest that the suspect object is an APT, the run-time classifier 850 may output a first score value. Depending on the weight assigned to scores provided by the run-time classifier 850 (as compared to other scores produced by analysis of the anomalous behaviors received from behavior monitoring logic 840 by score determination logic 860 and Score_1 provided by static analysis logic 810), the output first score value may significantly (and perhaps definitely) cause the overall score produced by score determination logic 860 to represent that the suspect object as malicious. Similarly, when determining that the IOCs suggest that the suspect object does not belong to any malware or APT family, the run-time classifier 850 may output a second score value less than the first score value. Again, depending on the weight assigned, the second score value may have little or no impact in assisting the score determination logic 860 to classify the suspect object as malicious.

According to one embodiment, it is contemplated that Score_2 output from the run-time classifier 150 may be based on the particular APT or malware family to which the suspect object belongs, where each classified malware and APT family is assigned as particular score value. Of course, it is contemplated that the score values simply may vary between types of families (APT, malware, etc.).

Referring now to FIG. 9, an exemplary block diagram of An illustrative embodiment of the communication system 900 implemented with electronic device 800 of FIG. 8 operating as a threat detection and prevention (TDP) system is shown. Herein, the communication system 900 comprises one or more TDP systems (e.g. TDP systems 910₁-910₃) coupled to management system 920 through a network 925. Herein, the management system 920 may be adapted to upload information associated with recently uncovered APTs and other malware into the TDP systems 910₁-910₃, such as newly updated malware family identifiers and APT family identifiers to database 160.

As shown, the TDP system 910₁ is adapted to analyze one or more objects associated with network traffic that may have originated from server device 932 via local network 930 and is now propagating over an enterprise network 934. The TDP system 910₁ is shown as being coupled with the local network 930, normally behind a firewall 936, via a network interface 938. The network interface 938 operates as a data capturing device (referred to as a “tap” or “network tap”) that is configured to receive network traffic propagating to/from the client device(s) 310A and provide object(s) from the network traffic to the TDP system 910₁.

In general, the network interface 938 is configured to receive and route one or more objects that are received from or targeted to client device 310A, normally without an appreciable decline in network performance. According to one embodiment of the disclosure, the network interface 938 may simply re-route an object for analysis to the TDP system 910₁ or, in another embodiment, duplicate the object and provide the same to the TDP system 910₁. For instance, the network interface 938 may duplicate one or more files that are part of a data flow or part of the payload contained within certain data packets, metadata, or the like.

It is contemplated that, for any embodiments where the TDP system 910₁ is implemented as an dedicated appliance or a dedicated computer system, the network interface 938 may include an assembly integrated into the appliance or computer system that includes network ports, network interface card and related logic (not shown) for connecting to the local network 930 to non-disruptively “tap” network traffic by providing at least a portion or a copy of the network traffic to TDP system 910₁. In other embodiments, the network interface 938 can be integrated into an intermediary device in the communication path (e.g., firewall, router, switch or other network device) or can be a standalone component.

Alternatively, the TDP system 910₁ may be deployed as an inline security appliance (not shown), which analyzes intercepted objects for malware or other indicators of suspicious content. Upon suspecting a presence of malware in an object under analysis, the suspect object may be forwarded to the dynamic analysis engine 970 for further analysis as described below.

More specifically, as shown in FIG. 9, the TDP system 910₁ comprises an I/O interface 945, static analysis logic 810, a first database 950 (along with family identifier database 160), a scheduler 960, a storage device 965, a dynamic analysis engine 970, object classifier 870 and reporting logic 880. In some embodiments, the network interface 938 may be contained within the TDP system 910₁ and operates as I/O interface 945. Also, the static analysis logic 810, the scheduler 960 and/or the dynamic analysis engine 970 may be software modules, which are executed by one or more processors (or different processors) and are configured to receive and analyze content within one or more received objects. After analysis, the potential APT objects (or TDP-detected features) are output from reporting logic 995 to client device 310A via I/O interface 945 and network interface 938.

In one embodiment, the static analysis logic 810 may serve as a filter to permit subsequent malware analysis only on a portion of incoming data, which effectively conserves system resources and provides faster response time in determining the presence of malware within the analyzed object(s). As shown in FIG. 9, the static analysis logic 810 is configured to receive incoming objects from the network interface 938 and applies heuristics to determine if any of the objects is “suspicious”. The heuristics applied by the static analysis logic 810 may be based on signature checks and/or rules stored in the database 955. Some of these rules may include APT-centric rules to uncover objects having certain traits common for APT malware (e.g., one or more unexpected attempts to exfiltrate data, etc.). Also, the static analysis logic 810 may examine the image of the object without executing or opening the object.

For example, the static analysis logic 810 may examine the metadata or attributes of the object under analysis (e.g., portion of an email message, file, document, a binary image of an executable, etc.) to determine whether a certain portion of the object statistically matches (e.g., a prescribed level of correlation with) a predetermined pattern of attributes that is associated with a malicious attack such as an APT attack. According to one embodiment of the disclosure, the static analysis logic 810 tags certain suspect objects within the network traffic as suspicious and supplies a score (Score_1 815) to score determination logic 860 for use in generating an overall score 865 for signaling to object classifier 870 as to whether the suspect object is malicious or not.

Thereafter, according to one embodiment of the invention, the static analysis logic 810 may be adapted to transmit the suspect objects to the dynamic analysis engine 970 and perhaps at least a portion of the metadata of the suspect objects to scheduler 960. The portion of the metadata may identify attributes of the runtime environment in which the suspect object should be processed and, on occasion, attributes of the client device(s) 310A to which the suspect object was targeted. Such metadata or attributes are used to identify a configuration of the VM needed for subsequent malware analysis. As an alternative embodiment, the dynamic analysis engine 970 may be adapted to receive one or more messages (e.g., data packets) from the static analysis logic 810 and analyze the message(s) to identify the software profile information associated with the needed VM and obtain such needed information.

As an illustrative example, a suspicious (suspect) object may constitute an email message that was generated, under control of Windows® 7 Operating System, using a Windows® Outlook 2010, version 1. Upon determining that the email message includes an attachment for example, static analysis logic 810 provides software profile information to the scheduler 960 to identify a particular configuration of VM needed to conduct dynamic analysis of the suspect object and its self-contained elements such as the attachment. According to this illustrative example, the software profile information would include (1) Windows® 7 Operating System (OS); (2) Windows® Outlook 2000, version 1; and perhaps an Adobe® reader if the attachment is a PDF document.

The static analysis logic 810 supplies the software profile information to the scheduler 960, which determines whether any of the VM disk files within storage device 965 feature a software profile supporting the above-identified configuration of OS and one or more applications or a suitable alternative.

The dynamic analysis engine 970 is adapted to execute one or more VMs 835₁-835_N, to simulate the receipt and execution of content associated with an object under analysis within a run-time virtual execution environment 830 as expected by the type of object. Furthermore, the behavior monitoring logic 840 within the dynamic analysis engine 970 may be configured to (i) monitor behaviors of the content being analyzed by one or more VMs 835₁, . . . , and/or 835_N, (ii) detect anomalous behaviors 845 associated with the monitored behaviors, and (iii) provide these anomalous behaviors 845 to both score determination logic 860 and run-time classifier 850. The run-time classifier 850 determines, through analysis of the anomalous behaviors (IOCs) and family identifiers (Common IOCs) as to whether there is a statistical match. If so, a score associated with the matched family identifier (Score_2 855) is provided to score determination logic 860.

Thereafter, based in part on Score_1 815, Score_2 855, and the results produced from analysis of the anomalous behaviors 845, the score determination logic 860 route the results (e.g., overall score 865, information associated with the detected anomalous behaviors, and other information associated with the detected malicious activity by the suspect object) to the object classifier 870.

According to one embodiment of the disclosure, the score determination logic 860 comprises one or more software modules that are used to determine a probability (or level of confidence) that the suspect object is malware. Score determination logic 860 is configured to generate the overall score 865 that classifies the threat of the possible malware. Of course, the overall score 865 may be based on a combination of different analysis results.

For instance, according to one embodiment, the overall score 865 may be an aggregation of a score independently generated by the score determination logic 860 along with Score_1 815 and Score_2 855. Alternatively, the overall score 865 may be an aggregation of these scores, with Score_2 855 being weighted more than Score_1 815. As another alternative, the overall score 865 may be weighted heavily on Score_2 855. In yet another embodiment, the overall score 865 may be based on a weighing primarily relying on the score produced by the score determination logic 860 separate and apart from Score_1 815 and Score_2 845, which may be used to assist in the analysis when the score produced by the score determination logic 860 is borderline as to whether the suspect object is malicious or not.

Referring to FIG. 10, an illustrative embodiment of a method for identifying malicious objects supplemented by use of a run-time classifier is shown. Herein, an object is received from a source such as a client device, a TDP system or the like (operation 1000). Upon receipt, the object undergoes static analysis to determine whether the object should undergo further in-depth analysis (e.g., virtual processing within a virtual execution environment) to better determine whether the suspect object is malicious (operation 1010). A first score indicative of a perceived threat level of the suspect object is generated and provided to a score determination logic.

If additional analysis is needed, the suspect object undergoes virtual processing to uncover anomalous behaviors associated with the suspect object in a sandboxed environment (operation 1020). Thereafter, a run-time analysis is conducted based on these anomalous behaviors (IOCs) and the family identifiers (Common IOCs), which represent known malware families (operation 1030). For instance, the run-time classifier may perform a statistical comparison between IOCs and Common IOCs).

Based on the analysis, a second score is output (operation 1040). Where the run-time classification determines that the suspect malware belongs to a classified (APT or non-APT) malware family, at least the name of the particular malware (APT or non-APT) family may be provided along with the second score.

Based on the uncovered anomalous behaviors, and taking in account at least the second score and perhaps the first score (along with any weighting applied to either of these scores), a determination is made as to whether the suspect object is malware (operations 1050). If so, the suspect object is identified as malware, and where the suspect object belongs to a known malware family, the name of the malware family (and other information associated with the identified malware family) may be provided (operations 1060-1070). Otherwise, the suspect object is identified as “benign” and the findings are reported.

Of course, it is contemplated that additional embodiments of the invention may be deployed. As a first example, logic components or the method of operation may be configured to determine whether the suspect object is malicious and also determine the type of malware (e.g. collective functionality of FIGS. 2 and 8). For instance, besides determining that the suspect object is malicious through an overall score as illustrated in FIG. 8, the run-time classifier may further assist the electronic device in determining (and subsequent reporting) whether the suspect object is an APT or a member of an APT family and the particulars of this determination. Other examples may include an embodiment of a first analysis may be conducted as to whether the suspect object includes any malware type (both APT and non-APT) and a subsequent, second analysis may be conducted to determine whether the malware is of a particular type (e.g. APT), or an embodiment where the malware family and APT family determination is conducted in a single operation in lieu of a series of operations, similar to the illustrative embodiment of FIG. 2.

In the foregoing description, the invention is described with reference to specific exemplary embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention as set forth in the appended claims.

Claims

1. A computerized method for identifying and classifying an object as belonging to a malware family, comprising: receiving one or more anomalous behaviors after processing of the object; anddetermining if the object is malware by performing a first analysis on the one or more anomalous behaviors and a pre-stored identifier identifying the malware family, the pre-stored identifier is a collection of data associated with anomalous behaviors that identify the malware family, the performing of the first analysis comprises determining a level of correlation between the one or more anomalous behaviors and the anomalous behaviors associated with the pre-stored identifier that are determined by (i) obtaining a plurality of anomalous behaviors, and (ii) removing one or more anomalous behaviors from the plurality of anomalous behaviors when the one or more anomalous behaviors exhibit (a) a first rate of occurrence in the malware family that is less than a first threshold and (b) a second rate of occurrence in one or more malware families other than the malware family that is greater than a second threshold to produce a subset of the plurality of anomalous behaviors that constitute the anomalous behaviors associated with the pre-stored identifier.

2. The computerized method of claim 1, wherein the malware family is an advanced persistent threat (APT) family.

3. The computerized method of claim 2, wherein the performing of the first analysis comprises performing a statistical comparison between the one or more anomalous behaviors and the anomalous behaviors associated with the pre-stored identifier that uniquely identify the APT family.

4. The computerized method of claim 3, wherein the performing of the first analysis further comprises determining that the one or more anomalous behaviors statistically match the anomalous behaviors associated with the pre-stored identifier that uniquely identify the APT family.

5. The computerized method of claim 1, wherein the removing of the one or more anomalous behaviors from the plurality of anomalous behaviors that exhibit the first rate of occurrence in the malware family comprises filtering at least one anomalous behavior having a count value less than a first count threshold from the plurality of anomalous behaviors to produce a first subset of anomalous behaviors, and the removing of the one or more anomalous behaviors from the plurality of anomalous behaviors that exhibit the second rate of occurrence in malware families other than the malware family comprises filtering at least one anomalous behavior having a count value greater than a second count value for a malware family other than the malware family from the first subset of anomalous behaviors to produce the subset of the plurality of anomalous behaviors, the subset of the plurality of anomalous behaviors being the pre-stored identifier.

6. The computerized method of claim 1, wherein the anomalous behaviors associated with the pre-stored identifier being a filtered subset of the plurality of anomalous behaviors associated with malware belonging to the malware family, the plurality of anomalous behaviors including the anomalous behaviors and at least one additional anomalous behavior different from any of the anomalous behaviors.

7. The computerized method of claim 1 further comprising performing a second analysis of the one or more anomalous behaviors subsequent to the first analysis upon failing to detect the level of correlation that includes conducting a statistical match after comparison of the one or more anomalous behaviors to any of a plurality of pre-stored identifiers each representing a different malware family, the plurality of pre-stored identifiers including the first pre-stored identifier.

8. The computerized method of claim 7, wherein the second analysis is conducted to determine if the object is a member of an advanced persistent threat (APT) family.

9. The computerized method of claim 8 further comprising: reporting results of the second analysis to a targeted destination, the results including information identifying one or more of an identifier for the APT family, a name of the APT family, or the one or more anomalous behaviors of the second analysis characteristic of the APT family.

10. The computerized method of claim 1, wherein the object is a flow comprising a plurality of related packets that are either received, transmitted, or exchanged during a communication session.

11. The computerized method of claim 1 further comprising: reporting results of the first analysis to a targeted destination, the results including information identifying one or more of (i) a family name of the malware family, (ii) the object, or (iii) the subset of the plurality of anomalous behaviors.

12. The computerized method of claim 11, wherein each of the subset of the plurality of anomalous behaviors is an indicator of compromise.

13. The computerized method of claim 1, wherein the pre-stored identifier includes a first plurality of indicators of compromise (IOCs) that are filtered from a second plurality of IOCs, where the first plurality of IOCs having a frequency of occurrence within the malware family substantially greater than an occurrence of any of the second plurality of IOCs excluding the first plurality of IOCs within the malware family.

14. The computerized method of claim 13, wherein the first plurality of IOCs are a combination unique to the malware family.

15. The computerized method of claim 1, wherein the first analysis on the one or more anomalous behaviors is conducted during run-time being a time that is contemporaneous with the processing of the object by one or more virtual machines and monitoring of the one or more anomalous behaviors.

16. An electronic device, comprising: a processor; anda memory communicatively coupled to the processor, the memory comprises virtual execution logic including at least one virtual machine configured to process content within an object under analysis and monitor for anomalous behaviors during the processing that are indicative of malware, andrun-time classifier logic that, when executed by the processor, performs a first analysis on the monitored anomalous behaviors and a pre-stored identifier to determine if the monitored anomalous behaviors indicate that the object is malware belonging to a classified malware family, the first analysis includes determining a level of correlation between the monitored anomalous behaviors and one or more anomalous behaviors associated with the pre-stored identifier that uniquely identify the classified malware family, the one or more anomalous behaviors being selected by (i) obtaining a first plurality of anomalous behaviors associated with malware belonging to the malware family, (ii) filtering at least one anomalous behavior having a count value less than a first count threshold from the first plurality of anomalous behaviors to produce a first subset of anomalous behaviors, (iii) filtering at least one anomalous behavior having a count value greater than a second count value for a malware family other than the malware family from the first subset of anomalous behaviors to produce a second subset of anomalous behaviors, the second subset of anomalous behaviors being the one or more anomalous behaviors associated with the pre-stored identifier.

17. The electronic device of claim 16, wherein the malware family is an advanced persistent threat (APT) family.

18. The electronic device of claim 17, wherein the first analysis performed by the run-time classifier logic comprises performing a statistical comparison between the monitored anomalous-behaviors and the anomalous behaviors associated with the pre-stored identifier that uniquely identify the APT family.

19. The electronic device of claim 16, wherein the first analysis performed by the run-time classifier logic comprises performing a statistical comparison between the monitored anomalous behaviors and the anomalous behaviors associated with the pre-stored identifier, the anomalous behaviors being a filtered subset of a plurality of anomalous behaviors associated with malware belonging to the malware family, the plurality of anomalous behaviors including the anomalous behaviors and at least one additional anomalous behavior different from any of the anomalous behaviors.

20. The electronic device of claim 16, wherein the determining of the level of correlation between the monitored anomalous behaviors and the one or more anomalous behaviors associated with the pre-stored identifier comprises performing a statistical comparison between the anomalous monitored behaviors and the one or more anomalous behaviors associated with the pre-stored identifier.

21. The electronic device of claim 16, wherein the memory further comprises a classifier that performs a second analysis of the monitored anomalous behaviors upon the run-time classifier failing to detect a statistical match upon comparing the monitored anomalous behaviors to any anomalous behaviors of a plurality of pre-stored identifiers each representing a different malware family, the plurality of pre-stored identifiers including the first pre-stored identifier.

22. The electronic device of claim 21, wherein the classifier conducts the second analysis by comparing the monitored anomalous behaviors operating as indicators of compromise to one or more anomalous behaviors associated with previously classified malware families.

23. The electronic device of claim 16, wherein the memory further comprises: reporting logic that, when executed by the processor, is configured to send results of the first analysis to a targeted destination, the results including information identifying one or more of (i) a family name of the malware family, (ii) the object, or (iii) the subset of the plurality of anomalous behaviors.

24. The electronic device of claim 23, wherein each of the subset of the plurality of anomalous behaviors is an indicator of compromise.

25. The electronic device of claim 16, wherein the memory further comprises: reporting logic that, when executed by the processor, is configured to send results of the second analysis to a targeted destination, the results including information identifying one or more of an identifier for an advanced persistent threat (APT) family, a name of the APT family, or the one or more anomalous behaviors of the second analysis characteristic of the APT family.

26. The electronic device of claim 16, wherein the first analysis on the monitored anomalous behaviors is conducted during run-time being a time that is contemporaneous with processing of the content within the object under analysis by the at least one virtual machine.

27. An electronic device, comprising: run-time classifier logic configured to perform a first analysis on (i) anomalous behaviors detected during processing of an object suspected of being malware and (ii) at least one pre-stored identifier being a collection of data associated with anomalous behaviors that uniquely identify a malware family, the first analysis to (1) determine if the anomalous behaviors indicate that the object is malware belonging to the malware family and (2) generate a score that represents a level of probability of the object being malware; andscore determination logic that is configured to use the score in determining whether the suspect object is to be classified as malware or not,wherein the anomalous behaviors of the at least one pre-stored identifier are selected by (i) obtaining a plurality of anomalous behaviors associated with malware belonging to the malware family, (ii) filtering at least one anomalous behavior from the plurality of anomalous behaviors when the at least one anomalous behavior has a count value less than a first count threshold to produce a first subset of anomalous behaviors, (iii) filtering at least one anomalous behavior from the first subset of anomalous behaviors when the at least one anomalous behavior has a count value greater than a second count value for a malware family other than the malware family to produce a second subset of anomalous behaviors, the second subset of anomalous behaviors being the anomalous behaviors associated with the at least one pre-stored identifier.

28. The electronic device of claim 27 further comprising: virtual execution logic including at least one virtual machine configured to process content within the object and monitors for the anomalous behaviors during processing of the object that are indicative of malware.

29. The electronic device of claim 28 further comprising: a static analysis engine adapted to (i) filter a first subset of objects having characteristics that are indicative of malware from a plurality of incoming objects and (ii) output the first subset of objects including the object to the virtual execution logic.

30. The electronic device of claim 27 further comprising: reporting logic configured to send results of the first analysis to a targeted destination, the results including information identifying one or more of (i) a family name of the malware family, (ii) the object, or (iii) the second subset of the plurality of anomalous behaviors.

31. The electronic device of claim 30, wherein each of the second subset of the plurality of anomalous behaviors is an indicator of compromise.

32. The electronic device of claim 27, wherein the first analysis on the anomalous behaviors is contemporaneous with the processing of the object by one or more virtual machines and monitoring of behaviors of the object during the processing of the object.