Cyber-security framework for application of virtual features

Abstract

A non-transitory storage medium having stored thereon logic wherein the logic is executable by one or more processors to perform operations is disclosed. The operations may include parsing an object, detecting one or more features of a predefined feature set, evaluating each feature-condition pairing of a virtual feature using the one or more values observed of each of the one or more detected features, determining whether results of the evaluation of one or more feature-condition pairings satisfies terms of the virtual feature, and responsive to determining the results of the evaluation satisfy the virtual feature, performing one or more of a static analysis to determine whether the object is associated with anomalous characteristics or a dynamic analysis on the object to determine whether the object is associated with anomalous behaviors.

Description

FIELD

Embodiments of the disclosure relate to the field of cyber security. More specifically, embodiments of the disclosure relate to a malware detection technique for performing an analysis of a group of objects, an object or a sub-object to determine whether further analysis should be performed on the group of objects, the object or the sub-object.

GENERAL BACKGROUND

Current network security appliances receive large amounts of data to analyze, possibly prior to passing to one or more endpoint devices. The large amounts of data make timely analysis difficult as analyzing the entire amount of received data is time-consuming and processing-intensive. Therefore, current network security appliances attempt to employ a filtering process that acts to determine whether portions of the received data exhibit some characteristics of suspiciousness prior to performing an in-depth analysis on the one or more portions of the received data.

However, current filtering techniques are reactive and attempt to match extracted features or other indicators of known malware with a portion of the received data in order to determine whether the portion of received data is suspicious. Specifically, malware detection systems within network security appliances often parse received data attempting to identify and then match a feature of the received data with an indicator of known malware to assess suspiciousness or even maliciousness. This technique often leads to false positives as benign (e.g., non-malicious) data may share an indicator with malware. As a result, in-depth malware analysis may be performed on the benign data, wasting time and requiring unnecessary processing. Of course, where the indicator is a fingerprint or signature in the form of a hash of known malware, the chance of a match with benign data is greatly reduced, but such a fingerprint or signature is often not available (e.g., as is the case with ‘zero day’ malware).

Many different malware detection approaches can suffer the effects of false positives. Known Intrusion Protection Systems (IPS) and Intrusion Detection Systems (IDS) are noteworthy for producing an enormous rate of false positives. Known “two-stage” advanced malware detection systems deployed at the periphery of a network employ a first static analysis stage used to filter incoming data before submitting that portion deemed suspicious for second-stage dynamic analysis via execution in a virtual environment. In such systems, the static analysis is configured to avoid false negatives and, as a trade-off, can unfortunately tag benign data as suspicious and then submit those false positives for dynamic analysis.

Additionally, received data may contain one or more features that seem benign in an isolated manner but, combined with other seemingly suspicious, malicious or benign features, are actually part of a malicious cyber-attack.

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:

Referring to FIG. 1, an exemplary block diagram of a threat detection system 110 including a real-time traffic analysis logic 111 deployed within the network 100 is shown.

FIG. 2 is an exemplary embodiment of a logical representation of the threat detection system of FIG. 1.

FIG. 3 is a flowchart illustrating an exemplary method for performing a real-time traffic analysis with the threat detection system of FIG. 1.

FIG. 4A is a sample illustration of a virtual feature.

FIG. 4B is a block diagram of a sample feature set from which the virtual feature of FIG. 4A is derived.

FIG. 4C is a sample table containing feature-condition pairings corresponding to the virtual feature of FIG. 4A.

FIG. 4D is a sample table containing the truth set corresponding to the virtual feature of FIG. 4A.

FIG. 5 is a block diagram of a sample table containing recorded observation values corresponding to the virtual feature of FIG. 4A and a sample object received by the threat detection system.

FIG. 6 is a block diagram of a sample table containing values representing feature-condition pairing evaluations corresponding to the virtual feature of FIG. 4A and the sample object received by the threat detection system of as discussed in FIG. 5.

FIG. 7 is a block diagram of a sample table containing the generated observation vector corresponding to feature-condition pairing evaluations as illustrated in FIG. 6.

DETAILED DESCRIPTION

Various embodiments of the disclosure relate to a real-time traffic analysis that determines whether a portion of received data should be provided to a static analysis logic and/or a dynamic analysis logic for an in-depth analysis. Such a real-time traffic analysis improves detection of exploits, malware attacks and/or anomalous characteristics/behaviors as the initial analysis performed by a real-time traffic analysis logic determines whether at least a portion of the received data has at least a first level of suspiciousness and/or maliciousness to warrant an in-depth analysis by either a static analysis logic and/or a dynamic analysis logic.

Real-time traffic analysis seeks to utilize extracted features of malware (which may themselves provide an indication that the portion of received data is suspicious and/or malicious), and/or seemingly benign features (which in isolation may not provide an indication that the portion of received data is suspicious and/or malicious) to create a more tailored, or directed, analysis. Real-time traffic analysis utilizes one or more “virtual features,” which may be each a combination of one or more features, typically expressed as a logical expression. Additionally, a condition may be applied to each feature within the virtual feature. A condition applied to a feature seeks to limit the values evaluated during real-time traffic analysis to those that, through experiential knowledge and/or machine learning, are known to be suspicious and/or malicious (the application of a condition to a feature may be referred to as a feature-condition pairing). Thus, the combination of feature-condition pairings within a logical expression (e.g., feature-condition pairings separated by logical operators comprising a virtual feature) leads to a more tailored analysis. The use of one or more virtual features to determine whether a portion of received data exhibits at least a first threshold of suspiciousness, maliciousness and/or anomalous characteristics and/or behaviors prior to performing an in-depth static and/or dynamic analysis, leads to fewer false positives. Thus, real-time traffic analysis enables a malware/threat detection system (e.g., deployed within a network appliance) to determine whether a portion of data exhibits at least a first threshold of suspiciousness, maliciousness and/or anomalous characteristics and/or behaviors more efficiently and accurately than pre-filter mechanisms that merely employ feature and/or signature matching techniques.

In one embodiment, the threat detection system receives an object to analyze. Subsequently, the object is passed to a real-time traffic analysis (RTTA) logic within the threat detection system. The RTTA logic parses the object looking for features of a predefined feature set. During parsing, the RTTA logic maintains a representation of the detected features of the predefined feature set as “observations,” and the value of each instance of the observation, if applicable. In one embodiment, the representation of values of detected features maintained by the RTTA logic may be a value for every instance of a detected feature (e.g., a size value for every embedded object) or a count of the number of instances of a feature (e.g., the number of JavaScript instances, embedded objects, Universal Record Locators (URLs) or other features that may appear multiple times within an object or portion of network traffic). In a second embodiment, the representation of detected features maintained by the RTTA logic may be a subset of the values of every instance of a detected feature. In such an embodiment, the RTTA logic may maintain the representation of detected features by eliding certain, similar values (e.g., omit duplicate values and/or merge similar values into ranges) of a detected feature. Upon completion of parsing, the representation of the values of detected features may be recorded in a storage device for accessing during an evaluation performed by the RTTA logic.

Once the object has been parsed and any values of observations have been recorded, the RTTA logic evaluates each feature-condition pairing set forth in the virtual feature against the value, or set of all values of the corresponding observation. As is defined below, the virtual feature is a combination of one or more feature-condition pairings, wherein the combination may be represented as a logical expression. Thus, in instances in which an observation has a value, the set of all values of an observation is paired with one or more conditions. In some embodiments, depending on the content of the received data, an observation may have a plurality of values (e.g., when observation_X, for example, is the size of an attachment to an email, observation_X will have a plurality of values when, for example, an email under analysis includes a plurality of attachments, wherein at least a first attachment has a different size than a second attachment).

Following the evaluation of each feature-condition pairing against the value, or set of all values of the corresponding observation, the RTTA logic generates an observation vector based on the feature-condition pairing evaluation and determines whether the observation vector is a member of the truth set corresponding to the virtual feature. A notification that the virtual feature was observed is provided to a RTTA score determination logic within the threat detection system. The threat detection system then generates a score indicating the level of suspiciousness and/or maliciousness of the portion of analyzed received data. In one embodiment, when the score is above a first threshold, the reporting logic may generate an alert that the object is malicious. When the score is greater than a second threshold but lower than the first threshold, the object may be provided to the static analysis logic and/or the dynamic analysis logic for further analysis. When the score is less than the second threshold, the threat detection system may determine no further analysis is needed (e.g., the object is benign).

The use of one or more virtual features by a threat detection system enables several advantages over current systems by providing a more efficient and flexible analysis. Specifically, the analysis performed by the RTTA using one or more virtual features enables the RTTA to combine observable features with conditions derived via experiential knowledge and/or machine learning to target particular exploits, vulnerabilities or malware. Additionally, an analysis including evaluating feature-condition pairings of a virtual feature using at least one or more values of a representative set of observed values within an object provides increased efficiency by eliminating the need to analyze all observed values using a rule-based detection scheme.

I. 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 are not limited or restricted to a microprocessor, one or more processor cores, a programmable gate array, a microcontroller, a controller, 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 link library, or one or more instructions. These software modules may be stored in any type of a suitable non-transitory (computer-readable) 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.

According to one embodiment, the term “malware” may be construed broadly as any code or activity that initiates a malicious attack and/or operations associated with anomalous or unwanted behavior. For instance, malware may correspond to a type of malicious computer code that executes an exploit to take advantage of a vulnerability, for example, to harm or co-opt operation of a network device or misappropriate, modify or delete data. Malware may also correspond to an exploit, namely information (e.g., executable code, data, command(s), etc.) that attempts to take advantage of a vulnerability in software and/or an action by a person gaining unauthorized access to one or more areas of a network device to cause the network device to experience undesirable or anomalous behaviors. The undesirable or anomalous behaviors may include a communication-based anomaly or an execution-based anomaly, which, for example, could (1) alter the functionality of an network device executing application software in an atypical manner (a file is opened by a first process where the file is configured to be opened by a second process and not the first process); (2) alter the functionality of the network device executing that application software without any malicious intent; and/or (3) provide unwanted functionality which may be generally acceptable in another context. Additionally, malware may be code that initiates unwanted behavior which may be, as one example, uploading a contact list to cloud storage without receiving permission from the user. For convenience, the terms “malware” and “exploit” shall be used interchangeably herein unless the context requires otherwise.

The term “processing” may include launching an application wherein launching should be interpreted as placing the application in an open state and performing simulations of actions typical of human interactions with the application. For example, the application, an Internet browsing application may be processed such that the application is opened and actions such as visiting a website, scrolling the website page, and activating a link from the website are performed (e.g., the performance of simulated human interactions).

The term “network device” may be construed as any electronic device with the capability of connecting to a network, downloading and installing mobile applications. Such a network may be a public network such as the Internet or a private network such as a wireless data telecommunication network, wide area network, a type of local area network (LAN), or a combination of networks. Examples of a network device may include, but are not limited or restricted to, a laptop, a mobile phone, a tablet, etc. Herein, the terms “network device,” “endpoint device,” and “mobile device” will be used interchangeably. The terms “mobile application” and “application” should be interpreted as software developed to run specifically on a mobile network device.

The term “malicious” may represent a probability (or level of confidence) that the object is associated with a malicious attack or known vulnerability. For instance, the probability may be based, at least in part, on (i) pattern matches; (ii) analyzed deviations in messaging practices set forth in applicable communication protocols (e.g., HTTP, TCP, etc.) and/or proprietary document specifications (e.g., Adobe PDF document specification); (iii) analyzed compliance with certain message formats established for the protocol (e.g., out-of-order commands); (iv) analyzed header or payload parameters to determine compliance, (v) attempts to communicate with external servers during processing in one or more VMs, (vi) attempts to access memory allocated to the application during virtual processing, and/or other factors that may evidence unwanted or malicious activity.

The term “virtual feature” may be interpreted as a combination of one or more feature-condition pairings, where the combination may be suspicious, malicious or unwanted. Additionally, a first virtual feature may be used as a feature within a second virtual feature. In one embodiment, the combination may be expressed as a logical expression having one or more “terms,” wherein a term may include one or more feature-condition pairings.

The term “feature” may be interpreted as a static trait that may be included in or exhibited by an object, group of objects or sub-object. Examples of features include, but are not limited or restricted to, a size of an object or sub-object, presence of an embedded object, size of an embedded object, number of embedded objects, presence of a URL in the object, number of URLs in the object, presence of a predefined signature, adherence or non-adherence to a predefined rule or protocol.

The term “observation” may be interpreted as features actually observed/detected within object, sub-object, etc. (for example, during or after parsing once the incoming traffic has been received).

The term “condition” may be interpreted as whether a predefined characteristic is present within the portion of network traffic being analyzed (e.g., a flow, a group of objects, an object, and/or a sub-object). Additionally, a condition may be a numeric evaluation such that a determination is made as to whether the set of all observed values of the feature include a particular number, intersect a given range, and/or contain any members greater than or less than a predefined threshold.

The term “observation vector” may be interpreted as a representation of the results of an evaluation of each condition with its corresponding feature, as evaluated for each feature-condition pairing of a single, virtual feature. In one embodiment, such a representation may take the form of a binary construct which may be understood as a vector or a one dimensional array. Herein, the term “observation vector” will be used but the use is not intended to limit the scope of the disclosure. In one example, a condition may be the presence of a feature. In a second example, a condition may be the size of the feature is greater than or equal to a predefined threshold. A condition may be applied to more than one feature (e.g., determining the presence of more than one feature) and/or a feature may be evaluated by more than one feature-condition pairing (e.g., a first term, as described below, of the virtual feature may determine whether a feature is greater than a first threshold and a second term of the virtual feature may determine whether the feature is less than or equal to a second threshold).

The term “truth set” may be interpreted as a representation of the one or more possible results of the evaluation of the feature-condition pairings of a virtual feature based on the values of the observations (e.g., detected features) that satisfy the terms of the corresponding virtual feature, where there is one truth set per virtual feature. The representation of the truth set may take the form of, for example, a set of binary vectors, wherein each binary vector represents a result of an evaluation of the feature-condition pairings of the virtual feature. All vectors included in the truth set represent solutions to the corresponding virtual feature.

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.” 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 by a cyber-security platform for performing a real-time traffic analysis of a group of objects, an object and/or a sub-object to determine whether the group of objects, the object and/or the sub-object is to be provided to a static analysis logic and/or a dynamic analysis logic for further analysis as to whether the group of objects, the object and/or the sub-object is suspicious, malicious or unwanted. 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.

II. Network Appliance Deployment

Referring to FIG. 1, an exemplary block diagram of a threat detection system 110 including a real-time traffic analysis logic 113 deployed within the network 100 is shown. In one embodiment, the network 100 may be an enterprise network that includes the threat detection system 110, a router 160, an optional firewall 161, a network switch 162, and one or endpoint devices 163. The network 100 may include a public network such as the Internet, a private network (e.g., a local area network “LAN”, wireless LAN, etc.), or a combination thereof. The router 160 serves to receive data, e.g., packets, transmitted via a wireless medium (e.g., a Wireless Local Area Network (WLAN) utilizing the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard) and/or a wired medium from the cloud computing services 170 and the endpoint devices 163. As is known in the art, the router 160 may provide access to the Internet for devices connected to the network 110.

In one embodiment, the network switch 162 may capture network traffic, make a copy of the network traffic, pass the network traffic to the appropriate endpoint device(s) 163 and pass the copy of the network traffic to the threat detection system 110. In a second embodiment, the network switch 162 may capture the network traffic and pass the network traffic to the threat detection system 110 for processing prior to passing the network traffic to the appropriate endpoint device(s) 163. In such an embodiment, the network traffic will only be passed to the appropriate endpoint device(s) 163 if the analysis of the network traffic does not indicate that the network traffic is associated with malicious, anomalous and/or unwanted characteristics and/or behaviors.

The threat detection system 110 includes a communication interface 111, a storage device 112, a real-time traffic analysis (RTTA) logic 113, a RTTA score determination logic 114, a static analysis logic 120, a dynamic analysis logic 130, a classification logic 140, and a reporting logic 150.

As shown, the threat detection system 110 is communicatively coupled with the cloud computing services 170, the Internet and one or more endpoint devices 163 via the communication interface 111, which directs at least a portion of the network traffic to the RTTA logic 113. The RTTA logic 113 includes one or more modules that, when executed by one or more processors, performs a real-time traffic analysis on the portion of the network traffic by analyzing the portion of the network traffic in light of one or more features and/or virtual features. When the real-time traffic analysis results in a score that is greater than a first threshold, the portion of the network traffic is provided to the static analysis logic 120 and/or the dynamic analysis 130. The static analysis logic 120 may include one or more software modules that, when executed by one or more processors, performs static scanning on a particular object, namely heuristics, exploit signature checks and/or vulnerability signature checks for example. The static analysis logic 120 and the dynamic analysis logic 130 may be one or more software modules executed by the same processor or different processors, where these different processors may be located within the same processor package (e.g., different processor cores) and/or located at remote or even geographically remote locations that are communicatively coupled (e.g., by a dedicated communication link) or a network.

More specifically, as shown, static analysis logic 120 may be configured with heuristics logic 121, exploit matching logic 122, and/or vulnerability matching logic 123. Heuristics logic 121 is adapted for analysis of certain portions of an object under analysis (e.g., the object may include a binary file) to determine whether any portion corresponds to either (i) a “suspicious” identifier such as either a particular Uniform Resource Locator “URL” that has previously been determined as being associated with known exploits, a particular source or destination (IP or MAC) address that has previously been determined as being associated with known exploits; or (ii) a particular exploit pattern. When deployed, the exploit matching logic 122 may be adapted to perform exploit signature checks, which may involve a comparison of an object under analysis against one or more pre-stored exploit signatures (e.g., pre-configured and predetermined attack patterns) from signatures database 124. Additionally or in the alternative, the static analysis logic 120 may be configured with vulnerability matching logic 123 that is adapted to perform vulnerability signature checks, namely a process of uncovering deviations in messaging practices set forth in applicable communication protocols (e.g., HTTP, TCP, etc.). The term “signature” designates an indicator of a set of characteristics and/or behaviors exhibited by one or more exploits that may not be unique to those exploit(s). Thus, a match of the signature may indicate to some level of probability, often well less than 100%, that an object constitutes an exploit. In some contexts, those of skill in the art have used the term “signature” as a unique identifier or “fingerprint,” for example, of a specific virus or virus family (or other exploit), which is generated for instance as a hash of its machine code, and that is a special sub-case for purposes of this disclosure.

The classification logic 140 may be configured to receive the static-based results (e.g., results from static analysis, metadata associated with the incoming network traffic, etc.), the VM-based results and/or the RTTA analysis results. According to one embodiment of the disclosure, the classification logic 140 comprises prioritization logic 141 and score determination logic 142. The prioritization logic 141 may be configured to apply weighting to results provided from the RTTA logic 113, the dynamic analysis logic 130 and/or the static analysis logic 120. Thereafter, the classification logic 140 may route the classification results comprising the weighting and/or prioritization applied to the RTTA results, the static-based results and/or the VM-based results to the reporting logic 150. The classification results may, among others, classify any malware and/or exploits detected into a family of malware and/or exploits, describe the malware and/or exploits and provide the metadata associated with any object(s) within which the malware and/or exploits were detected. Specifically, real-time traffic analysis using one or more virtual features may provide a tailored analysis such that each virtual feature may be directed to detecting characteristics typically representative of a particular exploit and/or type of attack (e.g., a ROP attack). Thus, use of the RTTA results may improve classification of a portion of network traffic (e.g., an object). The reporting logic 150 may generate an alert for one or more endpoint devices 163 and/or route the alert to a network administrator for further analysis. In addition, the reporting logic 150 may store the classification results (including the RTTA results, the static-based results and the VM-based results) in the storage device 112 for future reference. In one embodiment, results of the static analysis and/or the dynamic analysis may be provided to the RTTA logic 113 and/or the RTTA score determination logic 114. The RTTA logic 113 may evaluate one or more virtual features using the static analysis and/or the dynamic analysis results. The RTTA score determination logic 114 may utilize the results of the RTTA logic 113, the static analysis results and/or the dynamic analysis results to determine a score indicating a level of suspiciousness and/or maliciousness of the object. Additionally, the results of the evaluation performed by the RTTA logic 113 and/or the score determined by the RTTA score determination logic 114 may be provided to the classification logic 140 for use in classifying malware detected within the object.

When the static analysis logic 120 is provided at least a portion of the network traffic following a real-time traffic analysis by the RTTA logic 113, the static analysis logic 120 may perform a static analysis on the portion of network traffic. Subsequently, the static analysis logic 120 may route suspicious objects (and, in many cases, even previously classified malicious objects) to the dynamic analysis logic 130. In one embodiment, the dynamic analysis logic 130 is configured to provide, at least, an analysis of a binary included in the received network traffic and/or suspicious object(s) from the static analysis logic 120.

Upon receiving at least an object from the communication interface 111 and/or the static analysis logic 120, the dynamic analysis logic 130 performs processing within one or more VMs on the binary, e.g., the object is processed within the one or more VMs 131₁-131_j (where j≥1). The processing may occur within one or more virtual machine instances (VMs), which may be provisioned with a guest image associated with a prescribed software profile. Each guest image may include a software application and/or an operating system (OS). Each guest image may further include one or more monitors, namely software components that are configured to observe and capture run-time behavior of an object under analysis during processing within the virtual machine. During the processing within the virtual machine, the object is analyzed.

1. Logical Representation

FIG. 2 is an exemplary embodiment of a logical representation of the threat detection system of FIG. 1. The threat detection system 110 includes one or more processors 200 that are coupled to communication interface 111 via a first transmission medium 202. The communication interface 111, with the communication interface logic 211 located within a persistent storage 210, enables communication with network devices via the Internet, the cloud computing services and one or more the endpoint devices. According to one embodiment of the disclosure, the communication interface 111 may be implemented as a physical interface including one or more ports for wired connectors. Additionally, or in the alternative, the communication interface logic 111 may be implemented with one or more radio units for supporting wireless communications with other electronic devices.

The processor(s) 200 is further coupled to persistent storage 210 via a second transmission medium 203. According to one embodiment of the disclosure, persistent storage 210 may include (a) the static analysis logic 120 including a heuristics logic 121, an exploit matching logic 122, a vulnerability matching logic 123, and a signatures database 124, (b) a dynamic analysis logic 130 including one or more VMs 131₁-131_j, a virtual machine manager (VMM) 132, and a score determination logic 133, (c) a classification logic 140 including a prioritization logic 141, and a score determination logic 142, (d) a real-time traffic analysis (RTTA) logic 113, (e) a RTTA score determination logic 114, and (f) a reporting logic 150. Of course, when implemented as hardware, one or more of these logic units could be implemented separately from each other.

Additionally, although not shown, the RTTA logic 113 and/or the RTTA score determination logic 114 may be incorporated within any of the modules illustrated in FIG. 2. For example, the RTTA logic 113 and the RTTA score determination logic 114 may be incorporated within the static analysis logic 120. Furthermore, the RTTA logic 113 may include one or more modules. In one embodiment, the RTTA logic 113 includes a parsing logic, a feature detection logic (which maintains at least a representative set of observed values for one or more features), a feature-condition pairing evaluation logic (which may evaluate one or more feature-condition pairings using one or more values of a representative set of observed values, generate a result (e.g., an observation vector) and use the observation vector as an index into a truth set to determine whether one or more of the observed values satisfies the terms of the virtual feature.

III. Real-Time Traffic Analysis Methodology

Referring to FIG. 3, a flowchart illustrating an exemplary method for performing a real-time traffic analysis with the threat detection system of FIG. 1 is shown. Each block illustrated in FIG. 3 represents an operation performed in the method 300 of performing a real-time traffic analysis of one or more objects (or one or more sub-objects) by the real-time traffic analysis logic within a threat detection system. For convenience, in the description of FIG. 3, the term “object” will be used but the method and claims are not limited to such a scope. Referring to FIG. 3, at block 301, the communication interface of the threat detection system receives an object to analyze (optionally analyze on a per-sub-object basis). Subsequently, the object is passed to the real-time traffic analysis logic, which parses the object looking for features of a predefined feature set (block 302).

At block 303, during parsing, the RTTA logic maintains a representation of detected features of the predefined feature set as “observations.” For example, as illustrated in FIG. 5, the observations (e.g., detected features of the predefined feature set) may be recorded in a table. Alternatively, the observations may be stored in any representative form, examples include but are not limited or restricted to a linked list, hash table, vector or the like. Herein, for convenience, the term “table” will be used to refer to the representative form in which the observations are stored but is not intended to limited the scope of the disclosure.

Once the object has been parsed and any values of observations have been recorded, the RTTA logic evaluates each feature-condition pairing set forth in the virtual feature against the set of all the value(s) of the corresponding observation (block 304). As is defined above, the virtual feature is a combination of one or more feature-condition pairings, wherein the combination may be represented as a logical expression. Thus, each value of an observation is paired with one or more conditions. In some embodiments, depending on the content of the network traffic, an observation may have a plurality of values (e.g., when observation_X, for example, is the size of an attachment to an email, observation_X will have a plurality of values when the email under analysis includes a plurality of attachments, wherein at least a first attachment has a different size than a second attachment).

Following the evaluation of each feature-condition pairing for each value of the corresponding observation, the RTTA logic generates an observation vector based on the feature-condition pairing evaluation (block 305). The RTTA logic subsequently determines whether the observation vector is a member of the truth set corresponding to the virtual feature (block 306). In one embodiment, the determination of whether the observation vector is a member of the truth set may be done by using the observation vector as an index into a truth table representing the truth set (e.g., whether the observation vector corresponds to a value within the truth table). When the observation vector is a member of the truth set, mark the virtual feature as “observed” (block 307).

When the virtual feature is marked as observed (e.g., the observation vector is a member of the truth set corresponding to the virtual feature), a notification that the virtual feature was observed is provided to the RTTA score determination logic. When the score generated by the RTTA score determination logic, based on, inter alia, the observed virtual feature is greater than a predefined threshold, the object is provided to the static analysis logic and/or the dynamic analysis logic for further analysis (block 308).

Referring to FIG. 4A, a sample illustration of a virtual feature is shown. For convenience, a sample virtual feature, the virtual feature 400, is shown but does not limit the scope of the claims. A virtual feature may have more or less feature-condition pairings than the virtual feature 400. The feature-condition pairings as illustrated in virtual feature 400 are as follows: (i) Feature_A; (ii) Feature_N≥K₁; and (iii) Feature_N>K₂. As is illustrated, the features of virtual feature 400 are separated by logical operators. As discussed above, the use of one or more virtual features by a RTTA logic within a threat detection system enables the threat detection system to more efficiently and more accurately detect exploits, malware attacks and/or anomalous characteristics and/or behaviors as the virtual features provide a more tailored and directed detection. Thus, the RTTA logic more accurately determines on what data a more in-depth analysis should be performed.

Optionally, the results of the real-time traffic analysis may be provided to a classification logic to aid in classification of the object by potentially providing results of an analysis using the virtual feature that may be directed at detecting a specific exploit or type of attack.

Referring to FIG. 4B, a block diagram of a sample feature set from which the virtual feature of FIG. 4A is derived is shown. The virtual feature 400 is merely one embodiment of a virtual feature that may be derived from the feature set 410. The feature set 410 includes the one or more features as seen in FIG. 4B (in contrast to the feature-condition pairings as seen in FIG. 4C and mentioned above with respect to FIG. 4A). As mentioned above, for convenience, the sample virtual feature 400 is discussed herein; however, the scope of the claims is not so limited to the virtual feature 400, the feature set 410, the feature-condition pairings 420 of FIG. 4C, or the like.

The feature set 410 represents the features that the real-time traffic analysis logic will attempt to detect while parsing the object. Herein, the feature set 410 includes the two features as illustrated boxes 411 and 413. The feature of box 411 (Feature_A) is an embedded object, as illustrated in box 412. As will be shown below in FIG. 4C, Feature_A corresponds to a “presence” condition such that the RTTA logic will parse the object for the presence of an embedded object. Herein, the feature is not a specific type of embedded object, but merely any embedded object (however, a feature may be directed to a specific type). The feature of box 413 (Feature_N) represents a numeric feature. In contrast to the Feature_A that corresponds to the presence of a particular feature, Feature_N corresponds to a numeric observation. Herein, the sample Feature_N corresponds to the size of one or more embedded JavaScript instances.

Referring to FIG. 4C, a sample table containing feature-condition pairings corresponding to the virtual feature 400 of FIG. 4A is shown. As mentioned above, the feature-condition pairings 420 corresponding to the virtual feature 400 are set forth in the table of FIG. 4C. FIG. 4C. illustrates that a feature (herein, Feature_N) may correspond to more than one condition. Feature_N is shown to correspond to the feature-condition pairings illustrated in boxes 422 and 423 (Feature_N≥K₁, wherein K₁=200 kilobits (kb), and Feature_N>K₂, wherein K₂=300 kb, respectively). As will be discussed below, the evaluation of the observations will correspond to whether at least one observed value of Feature_N satisfies the conditions corresponding to the one or more feature-condition pairings.

Additionally, as is shown in box 421, a condition may be whether a feature (herein, Feature_A) has been observed. The example feature-condition pairing corresponding to Feature_A represents that the real-time traffic analysis logic will evaluate whether the presence of Feature_A (an embedded object) has been observed. In contrast to the numeric Feature_N, the number of embedded objects is not important, the feature-condition pairing is directed toward the presence of the feature, and thus, the number of instances of an embedded object greater than one does not change the outcome of the condition evaluation.

Referring to FIG. 4D, a sample table containing the truth set corresponding to the virtual feature of FIG. 4A is shown. As is defined above, a truth set is a set of binary vectors consisting of one or more vectors that satisfy the terms of the corresponding virtual feature. Thus, the vectors (e.g., the left-most three columns of the table 430) marked with a ‘1’ in the right-most column comprise the truth set corresponding to the virtual feature 400. The table 430 illustrates all possible vectors that may be derived from observations of the features, wherein the truth set is a subset of the table 430 consisting of {(0,0,1), (0, 1, 1), (1, 1, 0), (1, 1, 1)}.

Specifically, the left most column corresponds to the feature-condition pairing set forth in box 421 of FIG. 4C (the presence of Feature_A), the second left-most column corresponds to the feature-condition pairing set forth in box 422 of FIG. 4C (whether an observed instance of a value of Feature_N is greater than or equal to 2 Mb), and the third left-most column corresponds to the feature-condition pairing set forth in box 423 of FIG. 4C (whether an observed instance of a value of Feature_N is greater than 3 Mb). To explain further, a ‘0’ in the right-most column of table 430 corresponds to the feature-condition pairing not being satisfied and a ‘1’ corresponds to the satisfaction of the feature-condition pairing. As an example, the row 431 includes the boxes 432-434, which represent one possible vector that may be derived from the evaluation of the observations of the object. Specifically, the boxes 432-434 represent the vector wherein none of the feature-condition pairings were satisfied and the ‘0’ in box 435 represents that the vector comprised of boxes 432-434 {(0, 0, 0)} is not a member of the truth set corresponding to the virtual feature 400 (e.g., failing to satisfy any of the feature-condition pairings does not satisfy the terms of the virtual feature 400.

In one embodiment, the virtual feature 400, the feature set illustrated in table 410, the feature-condition pairings illustrated in table 420 and the truth set illustrated in table 430 are generated prior to analysis of the object. Therefore, in one embodiment, a network appliance has been preconfigured with the virtual feature 400, the feature set illustrated in table 410, the feature-condition pairings illustrated in table 420 and the truth set illustrated in table 430 prior to receipt of an object to analyze.

The receipt of an object for analysis by a threat detection system and the analysis of the object by, at least, the real-time traffic analysis logic within the threat detection system will be discussed below in conjunction with a discussion of FIGS. 5-7.

2. Parse Objects Looking for Features

Upon receiving network traffic (or alternatively, one or more objects, flows, group of objects, etc., may be received by the threat detection system in other manners such as via transfer from a storage device or the like), the communication interface provides at least a portion of the network traffic to the real-time traffic analysis (RTTA) logic. For convenience, the discussion herein will be directed to the analysis of an object, but the scope of the claims is not so limited.

The RTTA logic may parse the object on a per-object or per-sub-object feature basis. This enables the threat detection system to perform real-time traffic analysis at different granularities. A higher granularity may provide more detail and/or information to the static analysis logic and/or the dynamic analysis logic, when the object is passed on. However, performing RTTA at a high level of granularity (e.g., on a sub-object basis) may sometimes result in more processing for RTTA logic. A lower granularity may result in less processing for RTTA logic.

While parsing the object, the RTTA logic attempts to detect the features comprising the feature set (here, Feature_A and Feature_N). When the RTTA logic is evaluating object/sub-object for multiple virtual features, each virtual feature may use a different subset of the predefined feature set. Although not illustrated, the RTTA logic may be preconfigured with more than one virtual feature. In one embodiment, two or more virtual features may include the same subset of the feature set but the virtual features may be different (e.g., a first virtual feature may have a different order of feature-condition pairings separated by different logical operators than a second virtual feature).

3. Record Detected Features as “Observations”

Referring to a presence feature, such as Feature_A of FIG. 4B, the RTTA logic maintains a representation of whether the presence feature was detected during parsing, as an “observation.” For a presence feature, once the feature has been detected, the RTTA logic does not need to adjust the record of observations (as is illustrated in row 510 of FIG. 5). However, with respect to a numeric feature, such as Feature_N of FIG. 4B, a representation of the set of all observed values of the numeric feature is maintained, which may be recorded either immediately or upon the completion of parsing (the recordation is illustrated in row 520 of FIG. 5).

For example, a first feature may be a size of an embedded object and a second feature may be the number of JavaScript instances in an object. In such an example, referring to the first feature, an observation will correspond to a value for each instance of an embedded object and referring to the second feature, an observation will correspond to a value representing the number of JavaScript instances detected. Referring to a numeric feature, in one embodiment, the RTTA logic maintains the set of all observations corresponding to a feature, because a first observation may not satisfy all feature-condition pairings of the applicable feature but a second observation may satisfy one or more other feature-condition pairings of the applicable feature. Therefore, in order to evaluate the virtual feature using the entire object/sub-object, each observation (i.e., a value) of each detected instance of a numeric feature is to be recorded. Alternatively, in a second embodiment, the RTTA logic may maintain a representation of detected values corresponding to a feature that elides certain, similar values.

In one embodiment, a first numeric feature is the size of an embedded object, the RTTA logic maintains a representation of the set of all observations of embedded objects (a first value represented in the set corresponding to a first detected embedded object and a second value represented in the set corresponding to a second detected embedded object). Note that in such an example, the RTTA logic does not need to maintain duplicates (e.g., when a first embedded object and a second embedded object have the same size, a single entry in the representation of the set of all observations may be sufficient). When a second numeric feature is the number of, for example, JavaScript instances within a portion of network traffic, the RTTA maintains a representation (e.g., a count) of the number of JavaScript instances detected during parsing and records the final numeric value upon completion of the parsing.

Referring to FIG. 5, a block diagram of a sample table containing recorded observation values corresponding to the virtual feature of FIG. 4A and a sample object received by the threat detection system is shown. The values set forth as being observed in incoming traffic discussed herein, for example in FIGS. 5-7, are provided merely for convenience, but the scope of the claims is not necessarily so limited.

Herein, Observation_A corresponds to the observed value of Features and a set, Observation_N-1-Observation_N-3, corresponds to the observed values of Feature_N. According to FIG. 5, the received object for analysis did not include an embedded object, hence, the value of Observation_A is “Absent,” illustrated in in row 510. In contrast, FIG. 5 illustrates that three instances of embedded JavaScript were detected in the object, having values 160 kb, 220 kb and 310 kb, as shown in the set illustrated in row 520.

4. Evaluate Feature-Condition Pairings

As defined above, a “condition” may be either the presence of a feature or a numeric evaluation (e.g., above or below a threshold, within a specific numeric range, etc.). The evaluation of a feature-condition pairing includes determining whether the one or more detected values of an observation intersect with the subset of all possible values of that feature that satisfy the condition. The evaluation of each feature-condition pairing results in a “yes” or “no” answer. This may be represented in any manner such that there are two indicators, one representing an intersection and one representing no intersection (e.g., binary—‘1’ and ‘0’).

Referring to FIG. 6, a block diagram of a sample table containing values representing feature-condition pairing evaluations corresponding to the virtual feature of FIG. 4A and the sample object received by the threat detection system of as discussed in FIG. 5 is shown. As mentioned above, the values set forth as being observed in incoming traffic discussed herein are provided merely for convenience, but the scope of the claims is not necessarily so limited. Column 610 represents the features of the virtual feature 400, column 611 represents the observation values as discussed with respect to FIG. 5, and column 612 represents the result of evaluation of the feature-condition pairing using the observed values. As Observation_A has a value of “Absent,” (i.e., an embedded object was not detected within the object), the result of the evaluation of the feature-condition pairing is negative, or ‘0’ (illustrated in row 601). As the RTTA logic detected three values of Feature_N, feature-condition pairings involving Feature_N are evaluated against a set representation of those three values. Thus, illustrated in row 602, two observed values of Observation_N, 220 kb and 310 kb (shown via highlighting), result in a positive evaluation of the feature-condition pairing (Feature_N>200 kb), which results in a ‘1’ in box 625. Row 603 illustrates that only one value of ObservationN, 310 kb, results in a positive evaluation, or ‘1’ in box 628 (shown via highlighting in box 627).

5. Generate Observation Vector

The observation vector, as defined above, is a binary representation of the evaluation of the feature-condition pairings. Specifically, the observation vector contents are ordered according to ordering of the corresponding feature-condition pairing from the virtual feature.

Referring to FIG. 7, a block diagram of a sample table containing the generated observation vector corresponding to feature-condition pairing evaluations as illustrated in FIG. 6 is shown. As is illustrated in FIG. 6, the observation of table 700 corresponds to the results of the feature-condition pairings set forth in FIG. 5. Specifically, upon evaluating one or more values of the observations (e.g., detected features), the RTTA generates an observation vector by placing the results of the feature-condition pairings into a binary vector.

The observation vector of FIG. 6 was generated by the RTTA logic by placing the result of the feature-condition pairing evaluation of Feature_A (box 622) in the left-most column of table 700. Similarly, the RTTA logic places the result of the feature-condition pairing evaluation of Feature_N>200 kb (box 625) in the center column of table 700 and the result of the feature-condition pairing evaluation of Feature_N>300 kb (box 628) in the right-most column of table 700. Thus, the observation vector generated as a result of evaluating the feature-condition pairings using the observed values of the feature set corresponding to virtual feature 400 is {(0, 1, 1)}.

6. Determine Membership of Observation Vector in Truth Set

As defined above, the truth set corresponding to the virtual feature is a set of the one or more binary vectors that each satisfy the terms of the virtual feature. The determination as to whether the observation vector is a member of the truth set (and thus satisfies the terms of the virtual feature) is done by determining whether the observation vector matches a binary vector within the truth set.

Referring to FIGS. 4D and 7, once the RTTA logic has generated the observation vector, the RTTA logic determines whether the observation vector is a member of the truth set. As discussed above with respect to FIG. 4D, the truth set corresponding to the virtual feature 400 consists of {(0,0,1), (0, 1, 1), (1, 1, 0), (1, 1, 1)}. The RTTA logic determines whether the observation vector is a member of the truth set (i.e., whether a vector identical to the observation vector is found among the vectors comprising the truth set); when the RTTA logic determines a match between the observation vector and a vector within the truth set exists, the RTTA logic determines the observation set is a member of the truth set. As mentioned above, in one embodiment, the determination of whether the observation vector is a member of the truth set may be done by using the observation vector as an index into the truth table.

7. Mark Virtual Feature as “Observed” and Provide Observed Virtual Feature(s) to Score Determination Logic

Once the observation has been determined to be a member of the truth set, the virtual feature is marked as “observed.” The virtual feature may be marked as observed in the preconfigured feature set, making it available to subsequent virtual features, as well as to the RTTA score determination logic. Alternatively, the RTTA may mark which virtual features, if any, were observed in a separate table and pass the table to the RTTA score determination logic.

The RTTA score determination logic determines a score for the object that indicates a level of suspiciousness for the object. The score may be based on the one or more virtual features observed. In one embodiment, the score may also be based on the value(s) of one or more observations, and/or the presence or absence of one or more values. When the RTTA score determination logic determines a score that is above a predetermined threshold, the RTTA logic passes the object, or portion of network traffic that includes the object, to the static analysis logic and/or the dynamic analysis logic, as such a score indicates at least a first level of suspiciousness and/or malicious that warrants further analysis. In contrast, when the RTTA score determination logic determines a score that is not above the predetermined threshold, the RTTA logic does not provide the object, or portion of network traffic, to the static analysis logic or the dynamic analysis logic but may instead pass the object, or portion of the network traffic, on to one or more endpoint devices.

IV. Endpoint Device Deployment

In another embodiment, the threat detection system including real-time traffic analysis logic may be deployed in an endpoint device. Similar to the deployment discussed above regarding the network appliance, the threat detection system deployed within an endpoint device includes a real-time traffic analysis (RTTA) logic and a RTTA score determination logic. Data received by the endpoint device (e.g., via a network and/or data received via alternative manners, such as through a physical connection) is passed to the RTTA logic which parses the data and attempts to detect one or more features of a predefined feature set from which one or more virtual features are derived. The endpoint device may be pre-configured to include a threat detection system including the virtual feature(s), and the feature set (e.g., the threat detection system may be pre-installed on the endpoint device or may be installed on the endpoint device at any time prior to receipt of the data to be analyzed).

As discussed above, after detecting one or more features, and recording the representation of the values of each instance of that feature, if applicable, the RTTA logic evaluates the feature-condition pairings for a virtual feature using the observed/detected features (and values) and subsequently generates an observation vector based on the evaluation. The RTTA logic determines whether the observation vector is a member of the truth set that satisfies the terms of the virtual feature and, when the observation vector is a member of the truth set, an indication that the virtual feature was observed is passed to the RTTA score determination logic. The RTTA score determination logic determines a score indicating a level of suspiciousness, maliciousness and/or anomalous characteristics and/or behaviors. When the score is above a predetermined threshold, the threat detection system performs in-depth static and/or dynamic analysis on the portion of received data in order to determine whether the portion of data is suspicious, malicious or exhibits anomalous and/or unwanted characteristics and/or behaviors.

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 non-transitory storage medium having stored thereon logic, the logic being executable by one or more processors to perform operations including: generating a virtual feature including one or more feature-condition pairings represented by a logical expression, each feature-condition pairing expressed as a combination of one or more features along with a condition applied to each corresponding feature of the combination of the one or more features, wherein each condition limits the corresponding feature to a value that, through experiential knowledge or machine learning techniques, is known to be suspicious or malicious;detecting a plurality of features associated with an object, wherein each detected feature of the plurality of features constitutes a static characteristic included in or exhibited by the object;observing values associated with each of the plurality of features;performing a preliminary analysis by at least evaluating each observed value of the observed values associated with each of the plurality of features against the one or more feature-conditioned pairings of the virtual feature and generating an observation vector based on results of the preliminary analysis, wherein the results identify whether at least a first feature of the object having an observed value satisfies a condition of a feature-condition pairing of the one or more feature-condition pairings and the preliminary analysis is conducted to determine whether a static analysis or a dynamic analysis of the object is needed; andresponsive to the observation vector indicating that the virtual feature has been observed and the preliminary analysis determines that at least the one or more features associated with the object exhibits at least a first threshold of suspiciousness or maliciousness, performing one or more of (i) the static analysis to determine whether the object is associated with anomalous characteristics or (ii) the dynamic analysis to determine whether the object is associated with anomalous behaviors.

2. The non-transitory storage medium of claim 1, wherein preliminary analysis is configured to determine whether at least the first feature exhibits at least the first threshold of suspiciousness or maliciousness in response to the results of the preliminary analysis indicate at least the virtual feature has been observed in connection with the object.

3. The non-transitory storage medium of claim 2, wherein the results operate as an index for a representation of a truth set corresponding to the virtual feature.

4. The non-transitory storage medium of claim 1, wherein the one or more feature-condition pairings includes a group of features forming the virtual feature.

5. The non-transitory storage medium of claim 1, wherein the virtual feature is a combination of the one or more feature-condition pairings, the combination indicating a likelihood of the corresponding feature being an anomalous characteristic or an anomalous behavior based on the experiential knowledge or machine learning techniques.

6. The non-transitory storage medium of claim 5, wherein the anomalous characteristic includes a suspicious, malicious or unwanted characteristic or the anomalous behavior includes a suspicious, malicious or unwanted behavior.

7. The non-transitory storage medium of claim 1, wherein one or more observed values correspond to a size of an attachment to an email received as part of network traffic, the object being the attachment.

8. A computerized method comprising: generating a virtual feature including one or more feature-condition pairings represented by a logical expression, each feature-condition pairing expressed as a combination of one or more features along with a condition applied to each corresponding feature of the combination of the one or more features, wherein each condition limits the corresponding feature to a value that, through experiential knowledge or machine learning techniques, is known to be suspicious or malicious;detecting one or more features associated with an object, wherein each detected feature of the one or more features constitutes a static characteristic included in or exhibited by the object;observing values associated with each of the one or more features;performing a preliminary analysis in real-time by generating an observation vector constituting a representation of evaluations of each observed value of the one or more observed values against the one or more feature-condition pairings of the virtual feature, wherein the preliminary analysis is to determine whether further analyses are needed towards each corresponding feature with an observed value that satisfies a condition of a feature-condition pairing associated with the corresponding feature; andresponsive to the observation vector indicating that the virtual feature has been observed and the preliminary analysis determines that at least the one or more features associated with the object exhibits at least a first threshold of suspiciousness or maliciousness, performing one or more of (i) a static analysis to determine whether the object is associated with anomalous characteristics or (ii) a dynamic analysis to determine whether the object is associated with anomalous behaviors.

9. The method of claim 8, wherein the preliminary analysis is configured to determine whether at least the corresponding feature exhibits at least the first threshold of suspiciousness or maliciousness in response to results of the preliminary analysis indicate the virtual feature has been observed in connection with the object.

10. The method of claim 9, wherein the results operate as an index for a representation of a truth set corresponding to the virtual feature.

11. The method of claim 8, wherein the one or more feature-condition pairings includes a group of features forming the virtual feature.

12. The method of claim 8, wherein the virtual feature is a combination of the one or more feature-condition pairings, the combination indicating a likelihood of the corresponding feature being an anomalous characteristic or an anomalous behavior based on the experiential knowledge or machine learning techniques.

13. The method of claim 12, wherein the anomalous characteristic includes a suspicious, malicious or unwanted characteristic or the anomalous behavior includes a suspicious, malicious or unwanted behavior.

14. The method of claim 8, wherein one or more observed values correspond to a size of an attachment to an email received as part of network traffic, the object being the attachment.

15. The method of claim 8, wherein each feature included within the one or more feature-condition pairings of the virtual feature is set forth as a logical expression.

16. A system comprising: one or more processors; anda non-transitory, storage medium communicatively coupled to the one or more processors and having instructions stored thereon, the instructions, upon execution by the one or more processors, cause performance of operations including: parsing an object;detecting one or more features associated with an object, wherein each detected feature of the one or more features constitutes a static characteristic included in or exhibited by the object;observing one or more values associated with each of the one or more features;generating a virtual feature including one or more feature-condition pairings represented by a logical expression, each feature-condition pairing expressed as a combination of one or more features along with a condition applied to each corresponding feature of the combination of one or more features, wherein each the condition limits the corresponding feature to a value that, through experiential knowledge or machine learning techniques, is known to be suspicious or malicious;performing a preliminary analysis in real-time by conducting evaluations between each observed value of the one or more observed values against the one or more feature-condition pairings of the virtual feature, generating an observation vector constituting a representation of the evaluations, and directing further analyses toward each corresponding feature with an observed value that satisfies a condition of a feature-condition pairing associated with the corresponding feature; andresponsive to the observation vector indicating that the virtual feature has been observed and the preliminary analysis determines that at least the one or more features associated with the object exhibits at least a first threshold of suspiciousness or maliciousness, performing one or more of a static analysis to determine whether the object is associated with anomalous characteristics or a dynamic analysis to determine whether the object is associated with anomalous behaviors.

17. The system of claim 16, wherein the preliminary analysis is configured to determine whether at least the corresponding feature exhibits at least the first threshold of suspiciousness or maliciousness in response to results of the preliminary analysis indicate the virtual feature has been observed in connection with the object.

18. The system of claim 17, wherein the results operate as an index for a representation of a truth set corresponding to the virtual feature.

19. The system of claim 16, wherein the one or more feature-condition pairings includes a group of features forming the virtual feature.

20. The system of claim 16, wherein the virtual feature is a combination of the one or more feature-condition pairings, the combination indicating a likelihood of the corresponding feature being an anomalous characteristic or an anomalous behavior based on the experiential knowledge or machine learning techniques.

21. The system of claim 20, wherein the anomalous characteristic includes a suspicious, malicious or unwanted characteristic or the anomalous behavior includes a suspicious, malicious or unwanted behavior.

22. The system of claim 16, wherein the preliminary analysis generates an observation vector constituting a representation of the evaluation of each of the one or more observed values against the one or more feature-condition pairings.

23. The system of claim 16, wherein one or more observed values correspond to a size of an attachment to an email received as part of network traffic, the object being the attachment.