Patent Grant
11552969
Various embodiments concern computer programs and associated computer-implemented techniques for detecting email-based threats in the security field.
Employees of enterprise organizations (or simply “enterprises”) will often receive malicious email in their inboxes. Some of these malicious emails are quite sophisticated. For instance, a malicious email that constitutes an attack on the employee or the enterprise may be designed to bypass existing protective measures, reach the employee's inbox, and then be opened. Such emails usually arrive unbeknownst to the security team of the enterprise.
Many employees will take action(s) upon receiving a malicious email that place data (e.g., her own personal data or enterprise data) at risk. For example, an employee may click on a link embedded in the malicious email, provide her credentials, send confidential information, or transmit money to the unauthorized entity (also referred to as an “attacker” or “adversary”) responsible for generating the malicious email. Performance of such actions may result in the installation of malicious software, theft of credentials, compromise of the employee's email account, exfiltration of data, or theft of money.
Upon discovering the breach, the enterprise faces significant ramifications. These ramifications include:
Conventional email security software struggles to deal with attacks involving sophisticated malicious emails for several reasons.
First, an active adversary responsible for crafting personalized messages is often on the other side of sophisticated malicious emails. This is in contrast to less sophisticated email-based attacks, where a single person may send thousands or millions or generic, non-personalized emails in an attempt to succeed by sheer volume. Here, each sophisticated attack is new, unique, and personalized (e.g., for an employee or enterprise). Thus, employees will not observe the same sophisticated attack multiple times.
Second, sophisticated malicious emails generally will not include any attack signatures. The term “attack signature,” as used herein, refers to a trait that has been previously observed in one or more emails determined to be malicious. Conventional solutions typically rely on attack signatures as well as pattern matching, but sophisticated malicious emails can obviate these conventional solutions through personalization. Moreover, some sophisticated malicious emails do not contain any links or attachments. Instead, a sophisticated malicious email may simply contain text, such as “Hey, can you handle a task for me?” Upon receiving a response, the adversary may instruct the employee to, for example, wire money or share data. Further, in the case of an employee's email account, all emails will originate from the actual email account, thereby making it extremely difficult to detect malicious activity.
Third, the volume of emails handled by an enterprise is large, and receipt of emails is time sensitive. The decision on whether an email constitutes fraud should be rendered quickly for most emails as email security software should not inject delay into the flow of email. However, in most instances, conventional email security software indefinitely delays the delivery of email determined to represent a security threat.
Fourth, a relatively small number of sophisticated malicious emails will be handled by an enterprise over a given timeframe. For example, an enterprise may only observe several examples of sophisticated malicious emails over the course of a week. Accordingly, breaches due to sophisticated content are rare, and thus little data exists that can be ingested by machine learning (ML) models designed to identify sophisticated malicious emails.
Thus, there is a need in the security field to create computer programs and associated computer-implemented techniques for detecting email-based threats and then mitigating those threats.
A significant portion of targeted attacks on enterprises or their employees start with email, and these security threats are always evolving. As discussed above, a significant need for detecting and then resolving sophisticated email-based threats is becoming increasingly apparent. Conventional email security software does not adequately address the need for accurate, fast, and consistent detection of sophisticated malicious emails before those emails enter an inbox.
While a variety of different attack types should be addressed by a solution, two particular attack types present challenges in terms of detection and resolution. The first attack type is email account compromise (also referred to as “account takeover”). This form of attack is one in which an adversary accesses an employee's account with stolen credentials and then uses those credentials to steal money or data from the enterprise, or to send emails from the account in an effort to steal money or data from the enterprise or other employees. The second attack type is business email compromise. This form of attack is one in which an adversary impersonates an employee or a partner (e.g., a vendor). For example, the adversary may cause incoming email to appear as though it has been written by an employee (e.g., by changing the display name). This form of attack is usually performed in an effort to have an invoice—either legitimate or fictional—paid by the enterprise or to steal data.
Introduced here are threat detection platforms designed to collect and examine emails in order to identify security threats to an enterprise. A threat detection platform (also referred to as an “email security platform”) can be designed to address the above-mentioned attack types as well as other attack types such as phishing (e.g., campaign-based attacks), spear phishing (e.g., personalized attacks), extortion (e.g., cryptocurrency, gift card, and wire transfer ransoms), financial/data theft (e.g., vendor, partner, and client impersonation), and many other types of attacks, including those that have never been seen before.
At a high level, the technologies described herein can function to build a model representative of the normal email behavior of an enterprise (or an individual employee of the enterprise) and then look for deviations to identify abnormalities by applying the model to incoming emails. By establishing what constitutes normal behavior traits and/or normal email content, the enterprise can be protected against new, sophisticated attacks such as employee impersonation, vendor impersonation, fraudulent invoices, email account compromise, and account takeover. Moreover, canonicalizing, structuring, and storing data related to email may permit other high-value datasets to be created. For example, a threat detection platform may be able to derive valuable information regarding enterprise resource planning (ERP) from the email data. As further discussed below, the technologies described herein can leverage machine learning, heuristics, rules, human-in-the-loop feedback and labeling, or some other technique for detecting an attack (e.g., in real time or near real time) based on features extracted from a communication (e.g., an email) and/or context of the communication (e.g., recipient, sender, content, etc.).
Remediation action may be taken once a security threat has been detected. The remediation actions, if any, deemed appropriate may depend on the type of security threat that has been detected. For example, a threat detection platform may perform different remediation actions upon discovering a malicious email that includes an embedded link rather than a malicious email with an attachment. As part of the threat detection, identification, and remediation process, the threat detection platform may consider as input user actions; user-reported emails; machine learning (ML) training data including human-labeled emails, historical threat information, and scores; probabilities for threat detection based on models of known types of attacks; and heuristics including rules for blacklisting and/or whitelisting emails that meet certain criteria.
Various features of the technology will become more apparent to those skilled in the art from a study of the Detailed Description in conjunction with the drawings. Embodiments of the technology are illustrated by way of example and not limitation in the drawings, in which like references may indicate similar elements.
The drawings depict various embodiments for the purpose of illustration only. Those skilled in the art will recognize that alternative embodiments may be employed without departing from the principles of the technology. Accordingly, while specific embodiments are shown in the drawings, the technology is amenable to various modifications.
Introduced here are threat detection platforms designed to collect and examine emails in order to identify security threats to an enterprise. At a high level, the technologies described herein can function to build a model representative of the normal email behavior of an enterprise (or an individual employee of the enterprise) and then look for deviations to identify abnormalities by applying the model to incoming emails. By establishing what constitutes normal behavior traits and/or normal email content, the enterprise can be protected against new, sophisticated attacks such as employee impersonation, vendor impersonation, fraudulent invoices, email account compromise, and account takeover. As further discussed below, the technologies described herein can leverage machine learning, heuristics, rules, human-in-the-loop feedback and labeling, or some other technique for detecting an attack (e.g., in real time or near real time) based on features extracted from a communication (e.g., an email), attributes of the communication (e.g., recipient, sender, content, etc.), and/or datasets/information unrelated to communications. For instance, detecting sophisticated attacks that are plaguing an enterprise may require knowledge gleaned from multiple datasets. These datasets could include employee sign-in data, security events, calendar, contacts information, human resources (HR) information, etc. Each of these different datasets provides a different dimension to the normality of behavior of an employee and can be used to detect the most sophisticated attacks.
Remediation action may be taken once a security threat has been detected. The remediation actions, if any, deemed appropriate may depend on the type of security threat that has been detected. For example, a threat detection platform may perform different remediation actions upon discovering a malicious email that includes an embedded link rather than a malicious email with an attachment.
Embodiments may be described with reference to particular network configurations, attack types, etc. However, those skilled in the art will recognize that these features are equally applicable to other network configurations, attack types, etc. For example, while certain embodiments may be described in the context of the spear phishing attacks, the relevant feature(s) may be used in conjunction with other types of attacks.
Moreover, the technology can be embodied using special-purpose hardware (e.g., circuitry), programmable circuitry appropriately programmed with software and/or firmware, or a combination of special-purpose hardware and programmable circuitry. Accordingly, embodiments may include a machine-readable medium having instructions that may be used to program a computing device to perform a process for receiving input indicative of an approval to access email messages that were delivered to, or sent by, employees of an enterprise over a given interval of time, establishing a connection with a storage medium that includes the email messages, downloading the email messages into a local processing environment, building an ML model for identifying abnormal communication behaviors based on characteristics of the email messages, etc.
Terminology
References in this description to “an embodiment” or “one embodiment” means that the particular feature, function, structure, or characteristic being described is included in at least one embodiment. Occurrences of such phrases do not necessarily refer to the same embodiment, nor are they necessarily referring to alternative embodiments that are mutually exclusive of one another.
Unless the context clearly requires otherwise, the words “comprise” and “comprising” are to be construed in an inclusive sense rather than an exclusive or exhaustive sense (i.e., in the sense of “including but not limited to”). The terms “connected,” “coupled,” or any variant thereof is intended to include any connection or coupling between two or more elements, either direct or indirect. The coupling/connection can be physical, logical, or a combination thereof. For example, devices may be electrically or communicatively coupled to one another despite not sharing a physical connection.
The term “based on” is also to be construed in an inclusive sense rather than an exclusive or exhaustive sense. Thus, unless otherwise noted, the term “based on” is intended to mean “based at least in part on.”
The term “module” refers broadly to software components, hardware components, and/or firmware components. Modules are typically functional components that can generate useful data or other output(s) based on specified input(s). A module may be self-contained. A computer program may include one or more modules. Thus, a computer program may include multiple modules responsible for completing different tasks or a single module responsible for completing all tasks.
When used in reference to a list of multiple items, the word “or” is intended to cover all of the following interpretations: any of the items in the list, all of the items in the list, and any combination of items in the list.
The sequences of steps performed in any of the processes described here are exemplary. However, unless contrary to physical possibility, the steps may be performed in various sequences and combinations. For example, steps could be added to, or removed from, the processes described here. Similarly, steps could be replaced or reordered. Thus, descriptions of any processes are intended to be open-ended.
Technology Overview
Basic filtering services are offered by most email platforms.
Generally, the anti-spam filter 104 is designed to catch and quarantine malicious emails using blacklist(s) of senders, sender email addresses, and websites that have been detected in past unsolicited emails and/or policy frameworks defined by an enterprise. The term “anti-spam filter,” as used herein can refer to any legacy email security mechanism capable of filtering incoming emails, including secure email gateways (SEGs) (also referred to as “gateways”). For example, the enterprise (or the email service) may maintain a list of sender email addresses from which malicious email has been received in the past. As another example, an enterprise may decide to implement a policy that prohibits employees from receiving emails originating from a given domain. Malicious emails that are caught by the anti-spam filter 104 can be quarantined so as to remain hidden from the intended recipients, while non-malicious emails may be stored on an email server 106 (e.g., a cloud-based email server) for subsequent access by the intended recipients. Email servers (also referred to as “mail servers”) facilitate the delivery of emails from senders to recipients. Normally, an email will be transferred amongst a series of email servers as it travels toward its intended destination. This series of email servers allows emails to be sent between dissimilar email address domains.
Email servers can be broken down into two main categories: outgoing mail servers and incoming mail servers. Outgoing mail servers may be referred to as Simple Mail Transfer Protocol (SMTP) servers. Incoming mail servers will generally be either Post Office Protocol Version 3 (POP3) servers or Internet Message Access Protocol (IMAP) servers. POP3 servers are known for storing sent/received messages on local hard drives, while IMAP servers are known for storing copies of messages on servers (though most POP3 servers can store messages on servers too). Thus, the location of emails received by an enterprise may depend on the type of incoming mail server used by the enterprise.
As discussed above, such an arrangement is not suitable for recognizing sophisticated malicious emails. Accordingly, conventional filtering services often allow sophisticated malicious emails to reach employees' inboxes by mistake. Introduced here, therefore, are threat detection platforms designed to improve upon conventional filtering services.
Initially, the threat detection platform 214 may receive an email addressed to an employee of an enterprise. Upon receiving the email, the threat detection platform 214 may apply a first model 204 to the email to produce a first output indicative of whether the email is representative of a malicious email. The first model may be trained using past emails addressed to employee(s) of the enterprise that have been verified as non-malicious emails. Accordingly, the first model 204 may be referred to as the “surely-safe model.” The first model 204 serves as the first level of threat detection, and therefore may be tuned/designed to permit most email (e.g., upwards of 90, 95, or 99 percent of all incoming email) to reach the email server 206. Normally, the first model 204 is designed such that the initial threat determination is made fairly quickly (e.g., in less than 100, 50, or 25 milliseconds). Thus, the first model 204 may be responsible for performing load shedding.
If the email cannot be verified as non-malicious by the first model 204, the threat detection platform 214 can apply a second model 208 to the email. For the purpose of illustration, emails forwarded to the second model 204 may be referred to as “malicious emails.” However, these emails can be more accurately described as possibly malicious emails since the first model 204 is only able to verify whether emails are non-malicious. Upon being applied to the email, the second model 208 may produce a second output indicative of whether the email is representative of a given type of malicious email. Generally, the second model 208 is part of an ensemble of models applied to the email responsive to determining that the email is representative of a malicious email. Each model in the ensemble may be associated with a different type of security threat. For example, the ensemble may include separate models for determining whether the email includes a query for data/funds, a link to a Hypertext Markup Language (HTML) resource, an attachment, etc. As further discussed below, the second model 208 may be designed to establish different facets of the security threat responsive to a determination that the email is likely malicious. For instance, the second model 208 may discover facets of the security threat such as the strategy, goal, impersonated party, vector, and attacked party, and then upload this information to a profile associated with the intended recipient and/or the enterprise.
Then, the threat detection platform 214 may apply a third model 210 designed to convert the output produced by the second model 208 into a comprehensible visualization component 212. In embodiments where the second model 208 is part of an ensemble of models, the third model 210 can aggregate the outputs produced by the models in the ensemble, characterize the attack based on the aggregated outputs, and then convert the aggregated outputs into an explainable insight. For example, the third model 210 may generate a notification that identifies the type of security threat posed by the email, whether remediation actions are necessary, etc. As another example, the third model 210 may generate a human-readable insight (e.g., that includes text, graphics, or some combination thereof) using the facets, model features, and/or most discriminating features that triggered the determination that a security threat exists for the combination of email and attack. The explainable insight may be created so that security professionals responsible for addressing/mitigating security threats can more easily understand why the second model 208 flagged an incoming email as being representative of an attack.
Those skilled in the art will appreciate that the output of one model may be the entrance criteria for another model. Said another way, the order of the models employed by the threat detection platform 214 may triage which emails go to which models in an effort to reduce analysis time. Thus, the threat detection platform 214 may take a hierarchical bi-phasal approach to examining incoming emails.
The multi-tiered ensemble model may be comprised of different types of models, such as gradient boosting decision tree (GBDT) models, logistic regression models, and/or deep learning models. As further discussed below, each type of attack is generally scored by a separate model, so the threat detection platform 214 may employ different types of models based on the type of attack being detected.
Such an approach may be referred to as a “bi-phasal approach,” as it allows emails determined to be non-malicious to be routed to the email server 206 with very little delay while additional time is spent analyzing emails determined to be malicious (or at least potentially malicious).
Threat Detection Platform for Detecting Email-Based Threats
The system 300 can function to acquire email usage data of a customer (also referred to as an “enterprise”), generate a profile based on the email usage data that includes a number of received or inferred behavioral traits, monitor incoming emails, and, for each email, determine whether the email represents a security threat using a set of attack detectors (e.g., based on a deviation from behavioral trait(s) or normal content, such as by feeding the deviations into ML models), flag the email as a possible attack if the detectors indicate one, and if flagged, optionally perform one or more remediation steps on the email. The remediation step(s) may be performed in accordance with a customer-specified remediation policy and/or a default remediation policy. The term “customer,” as used herein, may refer to an organization (e.g., a corporation or an enterprise), a business unit, an individual (e.g., associated with one or more email addresses), a team, or any other suitable set of users of the threat detection platform 302. While embodiments may be described in the context of enterprises, those skilled in the art will recognize that the relevant technologies may be applied to other types of customers. As further discussed below, the system 300 may train one or more ML modules to serve as detectors capable of detecting a plurality of email attack types that can appear in the incoming email based on deviations from customer behavioral traits, normal email content, etc.
In some embodiments, the system 300 detects attacks based on the entire email (e.g., including the content of the body). However, in other embodiments, the system 300 is designed to detect attacks based only on the email metadata (e.g., information regarding the email header, sender, etc.) or some other suitable data.
All or portions of the system 300 can be implemented in an entity's email environment (e.g., the customer network 316), a remote computing system (e.g., through which incoming emails and/or data regarding incoming emails can be routed through for analysis), the entity's gateway, or another suitable location. The remote computing system can belong to, or be maintained by, the entity, a third-party system, or another suitable user. The system 300 may be integrated into the entity's email system: inline (e.g., at the secure email gateway), via an application programming interface (API) (e.g., wherein the system receives email data via an API such as the Microsoft Outlook® API), or another suitable manner. Thus, the system 300 can supplement and/or supplant other communication security systems employed by the entity.
In a first variation, the system 300 is maintained by a third party (also referred to as a “threat detection service”) that has access to multiple entities' emails. In this variation, the system 300 can route the emails, extracted features (e.g., primary attribute values), derivatory information (e.g., secondary attribute values), and/or other suitable communication information to a remote computing system maintained/managed by the third party. The remote computing system may be, for example, an instance on Amazon Web Services (AWS). In this variation, the system 300 can maintain one or more databases for each entity that include, for example, organizational charts, attribute baselines, etc. Additionally or alternatively, the system 300 may maintain federated database(s) shared amongst multiple entities such as detector databases, legitimate vendor databases, etc. In this variation, the third party can maintain different instances of the system 300 for different entities, or maintain a single instance for multiple entities. The data hosted in these instances can be obfuscated, encrypted, hashed, de-personalized (e.g., by removing personal identifying information (PII)), or otherwise secured or secreted.
In a second variation, the system is maintained by the entity whose emails are being monitored (e.g., remotely or on premises), and all data can be hosted by the entity's computing system. In this variation, data to be shared across multiple entities, such as detector database updates and new attack signatures, can be shared with a remote computing system maintained by a third party. This data can be obfuscated, encrypted, hashed, de-personalized (e.g., by removing PII), or otherwise secured or secreted. However, the system 300 can be maintained or executed using any other suitable computing and ownership configuration.
As shown in
The customer network 316 can be an enterprise network, mobile network, wired network, wireless network, wireless spectrum network, or any other communications network maintained by a customer or a network operator associated with the customer. As noted above, the customer can be an individual, an enterprise, or another suitable entity. For example, an enterprise may utilize the services of a computer security company for at least email threat detection. The enterprise may grant permission to the computer security company to monitor the customer network 316, including monitoring incoming emails at the customer network 316, analyzing the emails for potential threats, and performing some remediation action(s) if threats are detected. In some embodiments, the enterprise further grants permission to the computer security company to collect or receive various pieces of data about the enterprise in order to build a profile that specifies enterprise norms, behavioral traits, and normal email content.
The threat detection datastore 310 may include one or more databases in which customer data, threat analysis data, remediation policy information, customer behavioral traits or norms, normal customer email content, and other pieces of data can be stored. The data can be: determined by the system 300 (e.g., calculated or learned from data retrieved, received, or otherwise collected from the customer network 316 or the entity's email provider), received from a user, retrieved from an external database (e.g., LinkedIn® or Microsoft Office 365®), or otherwise determined. In some embodiments, the threat detection database 310 also stores output from the threat detection platform 302, including human-readable information on detected threats and actions taken. Other various data or entities may be stored.
Customer data can include, but is not limited to, email usage data; organizational data such as organization members/employees and their titles; customer behavior traits or norms (e.g., determined based on historical emails); attack history (e.g., determined based on historical emails, determined by applying the attribute extractors and/or analysis modules to historical emails, etc.); entity profile(s); normal customer email content; email addresses and/or phone numbers of organization members; identities of entities and/or individuals who organization members frequently communicate with, both internally and externally; email volume during various times of the day; topics or subject matter most frequently discussed, and how frequently; and more.
The system 300 can optionally include a profile generator 304 that generates one or more entity profiles (also referred to as “customer profiles”) based on past emails and/or email usage data associated with the entity. In a second variation, the system 300 includes multiple profile generators 304 that each extract one or more attributes of the entity profile. However, the system 300 can include any suitable number of profile generators in any suitable configuration.
Entity profiles can be generated for: each customer, each business unit, each individual (e.g., each employee or recipient of emails), each email address, each organization type, or another suitable entity or entity group. The entity profile is preferably used as the baseline for entity communication behavior (e.g., email behavior) but could be used in other manners. Moreover, profiles could be generated external to an entity, and these profiles may be federated across a customer base for use by all entities whose emails are being monitored by the system 300. For example, a profile for a trusted third party (e.g., Oracle), a representative of a trusted third party (e.g., a sales representative at Oracle), or a financial institution (e.g., with known routing numbers so as to detect fraudulent invoice payments) could be federated across a customer base. Thus, the system 300 may build a federated network of profiles modeling businesses, vendors, customers, or people.
An entity profile can include: primary attributes, secondary attributes, or any other suitable feature. These values can be: median values, mean values, standard deviations, ranges, thresholds, or any other suitable set of values (e.g., for the entity profile, extracted from new emails, etc.). The entity profile can include: time series (e.g., trends or values for a specific recurrent time, such as months of the year), static values, or may have other suitable contextual dependency.
Primary attributes are preferably attributes or features extracted directly from a communication, but could be otherwise determined. The primary attributes can be extracted by one or more primary attribute extractors, each extracting one or more primary attributes from the communication as shown in
Secondary attributes are preferably attributes that are determined from the primary attributes and/or customer data (e.g., as determined from the threat detection datastore 310), but can be otherwise determined. The secondary attributes can be extracted, inferred, calculated, or otherwise determined. The secondary attributes may be determined by one or more secondary attribute extractors, each extracting one or more secondary attributes from the primary attributes for a given communication or entity as shown in
The entity profile can additionally or alternatively include: a number of customer behavioral traits or typical email content associated with the customer. In some embodiments, the profile generator 304 receives the email usage data from the customer network 316 or the threat detection datastore 310. The email usage data can include, but is not limited to, information on email addresses of employees and contacts, email content (e.g., the body of email messages), frequency of email, volume of email at given times of day, HTML/font/style usage within email, confidential topics and explicitly or implicitly authorized members discussing those topics, spam mail and characteristics thereof, and more.
The entity profile can be generated from: historic email data for the entity (e.g., retrieved using an API to the entity's email environment, retrieved from an email datastore, etc.); newly-received email (e.g., email received after system connection to the entity's email environment); user inputs; other entities (e.g., sharing a common feature or characteristic with the entity); or based on any other suitable data. In some embodiments, one or more pieces of email usage data can be collected, generated, or inferred by the profile generator 304 based on received pieces of customer data, monitoring of the customer network 316 given authentication and access by the customer, or some other means.
The entity profile can be generated using the same system(s) as typical email analysis (e.g., using the attribute extractors used to extract the attributes for real- or near-real-time threat detection), but can alternatively or additionally be generated using other suitable system(s).
In one variation, the profile generator 304 generates a customer profile by building a deep profile of the company's email usage, membership roles and/or hierarchy, daily norms, behavioral traits, and more in order to establish a model of what “normal” or “typical” looks like for the customer in terms of email usage and behavior, and, by extension and inference, what “abnormal” or “atypical” emails and/or activity may constitute for purposes of identifying likely threats.
In some embodiments, the customer profile is generated based on received, collected, and/or inferred customer data, email usage data, and other relevant information. Examples of questions that the customer profile may seek to model answers for regarding the customer include, but are not limited to: What are normal email addresses for each member of the organization? What are normal topics for each person, pair, and/or department to be discussing (e.g., Joe and Samantha normally discuss product release plans, but never accounting or billing topics)? What are normal login or email sending times for each user? What Internet Protocol (IP) address do they log in from? What geolocation do users usually log in from? Do users have suspicious mail filter rules set up (e.g., attackers who hijack email accounts sometimes set up mail filters to auto-delete emails containing certain keywords in order to conceal illicit activity from the account's true owner)? What's the normal tone or style each user uses? What's the tone used between each pair of users? What's the normal signature (e.g., “cheers” or “thanks”) used by each employee? What types of words are used more in one department and less in another department? What are the normal vendors/partners that a customer communicates with and/or is billed by? Does a given pair of users usually talk? What are typical email authentication statuses for this person, pair, or entity (e.g., SPF, DKIM, or Domain-Based Message Authentication, Reporting, and Conformance (DMARC))? When a user receives or sends links/attachments, does a derived description of the link/attachment match a given description of the link/attachment? When an employee receives emails with attachments, what are the typical characteristics of the attachments (e.g., name, extension, type, size)?
The monitoring module 308 operates to monitor incoming emails at a network maintained by the customer. In some embodiments, the monitoring module 308 monitors incoming emails in real time or substantially real time. In some embodiments, the monitoring module 308 is authorized to monitor the incoming emails only upon the system 300 and/or the threat detection platform 302 being authenticated and granted permission and access by the customer network 316. In some embodiments, the system 300 and/or the threat detection platform 302 is integrated into an office suite or an email suite via an API.
The analysis module 312 operates to analyze each incoming email for possible threats/attacks. The analysis module 312 preferably detects attacks based on the secondary attributes (e.g., for one or more communications for the entity), but can alternatively or additionally detect attacks based on the primary attributes or any other suitable data. In one variation, the analysis module 312 is separate from the primary and secondary attribute extractor(s). In another variation, the analysis module 312 can include the primary and/or secondary attribute extractor(s). However, the analysis module 312 can be otherwise configured.
The system 300 can include one or more analysis modules 312, operating in parallel, in series, or in another suitable order. An example of multiple analysis modules 312 operating in conjunction with each other is shown in
However, the system 300 can include any suitable number of analysis modules 312 for detecting any number of attack types. In particular, the approach of modeling employed by the system 300 to discover behavioral norms of employee, vendor, and organizational communication allows for the ability to identify attacks that have not been seen before, as well as zero-day phishing attacks.
The analysis modules 312 can include or use one or more of: heuristics, neural networks, rules, decision trees (e.g., gradient-boosted decision trees), ML-trained algorithms (e.g., decision trees, logistic regression, linear regression, etc.), or any other suitable analysis method/algorithm. The analysis modules 312 can output: discrete or continuous outputs, such as a probability (e.g., attack probability), a binary output (e.g., attack/not attack), an attack classification (e.g., classification as one of a plurality of possible attacks), or output any other suitable output. The analysis modules 312 can be: received from a database (e.g., a database of known attack patterns or fingerprints), received from a user, learned (e.g., based on data shared across multiple entities, based on communication data for the entity, etc.), or otherwise determined.
Each analysis module can be specific to an attack, an attack type/class, or any other suitable set of attacks. The system 300 can include one or more analysis modules 312 per attack set. In one variation, the attack set can be associated with multiple analysis modules 312, where the system 300 can dynamically select the analysis module to use (and/or which output to use) based on the performance metrics of each analysis module for the given attack set. For example, the system 300 can include a heuristics-based analysis module and an ML-based analysis module for a given attack that are executed in parallel for each piece of communication; monitor the recall and/or the precision (e.g., as determined based on entity feedback of the email classifications) for both analysis modules; and select the analysis module with the higher performance value for subsequent communication analysis. The output of all but the highest-performant analysis module can be hidden from the user and/or not used in email attack classification; alternatively, the outputs of lower-performant analysis modules can be used to verify the highest-performant analysis module output, or otherwise used.
One or more of the analysis modules 312 can be specific to an entity (e.g., an organization, a business organization, a title, an individual, an email address, etc.), be shared between multiple entities (e.g., be a global analysis module), or be otherwise customized or generic.
In one example, first, for each incoming email, the analysis module 312 (e.g., secondary attribute extractor) determines a deviation of the mail from each of the plurality of customer behavioral traits or content norms. In some embodiments, the deviation is a numerical value or percentage representing a delta between the customer behavioral trait and an assigned behavioral trait determined from the incoming email. For example, if a customer behavioral trait is “Joe Smith almost exclusively sends email from js@customerentity.com” and an incoming email purporting to be from Joe Smith has the email address joesmith@genericmail.com, then the deviation will be assigned a high number. If Joe Smith sends from a genericmail.com account approximately 20 percent of the time, then the deviation will still be relatively high, but the deviation will be lower than the previous example. Second, the analysis module 312 feeds the measured deviations as input into one or more attack detectors (e.g., rules-based engines, heuristics engines, ML models, etc.), each of which can generate an output. Third, the analysis module 312 flags the email as a possible attack corresponding to an email attack type if an indication is received from one or more of the ML models that a deviation threshold for the email attack type has been exceeded. The analysis module 312 can instruct the ML models to categorize the deviations in an incoming email as indicating a likely malicious email or a likely non-malicious email, as well as categorize the email according to a possible attack type. In some embodiments, the ML model “trips”—that is, f(email) exceeds a threshold for deviations from customer behavioral traits and content norms—and then flags the email as a possible attack.
The output(s) produced by the analysis module(s) 312 can optionally be fed into a master detector that analyzes these output(s) in order to produce a final classification for the communication as an attack or non-attack, as shown in
The remediation engine 314 optionally operates to perform one or more remediation processes. The remediation engine 314 is preferably implemented in response to communication classification as an attack (e.g., by one or more analysis modules 312, by the master detector, etc.), but can alternatively or additionally be implemented at any other suitable time. In some embodiments, the remediation steps are based on or correlate to a customer remediation policy. The customer remediation policy may be predefined and received by the threat detection platform 302, be generated based on inference, analysis, and customer data, or be otherwise determined. In some embodiments, the threat detection platform 302 may prompt the customer to provide one or more remediation steps or components of the remediation policy in various situations. Remediation steps may include, for example, moving the email to a junk folder as shown in
In some embodiments, the remediation engine 314 provides threat detection results and/or other output to the customer via, for example, a customer device 318. Examples of customer devices 318 include mobile phones, laptop computers, and other computing devices. In some embodiments, the remediation engine 314 sends the output in a human-readable format to the threat detection platform 302 for display on an interface.
The system 300 can optionally include a training module 306 that operates to train the ML model(s) employed by the analysis module(s) 312. Each ML model preferably detects a single attack type, but can alternatively detect multiple attack types. In some embodiments, the training module 306 trains the ML models by feeding training data into the ML models. The training data can include: entity-labeled communications (e.g., system-analyzed emails that have been sent to security personnel and labeled as attacks or non-attacks, as shown in
In some embodiments, the training data for training the ML models can include human inputs received from the customer. Organizations often have phishing mailboxes where employees can report emails to security teams, or where security teams can automatically/manually reroute messages meeting certain criteria. The training data can include emails that are placed in these phishing mailboxes as malicious emails. In some embodiments, the human inputs can include end user actions that can feed into ML models. For example, if an individual moves an email that the ML models cannot decide with certainty whether to discard, that user action can be included as training data to train the ML models as to what action should be taken in similar contexts.
Examples of potential attack types that the ML models can be trained to detect in varying embodiments include, but are not limited to, vendor impersonation and ransom attacks.
In some embodiments, a plurality of heuristics data is utilized as an alternative to, or in combination with, the ML models to detect threats, train the ML models, infer behavioral traits or content norms for a customer based on customer data, select potential attack types of a customer, or perform other threat detection tasks. In some embodiments, one or more aspects of training the ML models includes feeding the plurality of heuristics data as input training data into one or more of the ML models. In some embodiments, the heuristics data is utilized in relation to a rules engine which operates to decide which heuristics to apply in different situations. In some embodiments, the rules engine determines whether to apply machine learning or heuristics in a particular threat detection task. In some embodiments, one or more rules may include a blacklist and/or a whitelist for certain email criteria.
In some embodiments, any level of granularity can be contemplated for the system 300 regarding the analysis module 312, ML models, heuristics, rules, and/or human labeling of inputs. In some embodiments, “normal” and “abnormal” behavioral traits and content norms could be determined on a per-employee, per-pair, per-department, per-company, and/or per-industry basis.
In some embodiments, the ML models can optionally be refined in a number of ways in the course of operation. In some embodiments, the monitoring module 308 monitors the customer's phishing mailbox to locate false negatives (i.e., emails that were missed by the ML models that employees subsequently reported to the security team). In some embodiments, a customer may reverse the remediation decision made by the heuristics and/or the ML models, and in response the ML models may incorporate that feedback. In some embodiments, if a customer marks a particular feature in an email (e.g., sender mail, display name, authentication state, etc.) as suspicious, that can feed back into the ML models. In some embodiments, such feedback is weighted in an ML model based on the stature or reputation of the individual responsible for providing the feedback. For example, an ML model may trust a Tier 3 employee's judgment on an email significantly more than a Tier 1 employee's judgment, and would weigh their feedback more heavily into the ML model.
In some embodiments, different types of ML models may be used, including but not limited to gradient-boosted decision trees, logistic regression, linear regression, and more. In some embodiments, ML models are replaced with purely rules-based engines.
The process 400 functions to provide email-based threat detection according to a generated customer profile that models normal customer behavior and normal email content, and then feeds the deviations from these normal behavior traits and normal content as input into ML models training on malicious emails.
In some embodiments, the process 400 is enabled by a web-based platform (e.g., threat detection platform 302 of
The process 400 can be performed in parallel or in series with email delivery to an email inbox. In one variation, the process 400 is performed in parallel with email delivery to the recipient's email inbox, wherein the email is retroactively removed from the email inbox in response to determination that the email is an attack (and/or has a high probability of being an attack). In a second variation, the process 400 is performed in series with email delivery, where the email is only delivered to the recipient's email inbox in response to determination that the email is not an attack. However, the process 400 can be otherwise integrated into email delivery paradigms. The method can analyze: all incoming emails, only emails labeled as non-attacks by a preceding security system, only emails labeled as attacks by a preceding security system, or any suitable set of emails.
Monitoring incoming emails (step 404) is preferably performed using a monitoring module (e.g., monitoring module 308 of
Email attributes are preferably determined by extractors, as discussed above, but can be otherwise determined. In one example, the method includes: extracting primary attributes from the incoming email (e.g., with one or more specialized primary attribute extractors executing in parallel), and determining secondary attributes for the email (e.g., with one or more specialized secondary attribute extractors executing in parallel) from the primary attributes and customer data.
The attack is preferably determined using one or more analysis modules, as discussed above, but can be otherwise determined. In one variation, the determined attributes (e.g., primary or secondary attributes) can be fed into one or more analysis modules (e.g., executing in parallel or in series). In some embodiments, each analysis module is specific to an attack type, where the plurality of outputs from the plurality of analysis modules is further analyzed (e.g., by a master detector) to determine whether the email is an attack. In other embodiments, the analysis module detects multiple attack types (e.g., outputs multiple output values, each corresponding to a different attack type, where the output can be a likelihood and/or confidence in the corresponding attack type), and the email can be labeled as an attack when the output value exceeds a predetermined threshold for the corresponding attack type. However, the attack can be otherwise detected.
Step 408 optionally includes performing remediation steps, as described above with respect to remediation engine 314 of
Step 401 includes collecting or receiving email usage data, as described above with respect to profile generator 304 of
Step 402 includes generating a customer profile based on the email usage data, as described above with respect to profile generator 304 of
Step 403 includes training ML models to detect email attack types, as described above with respect to training module 306 of
Step 405 includes measuring deviations in incoming email, as described above with respect to analysis module 312 of
Step 406 includes feeding the measured deviations into ML models, as described above with respect to analysis module 312 of
Step 407 optionally includes flagging email as a possible attack, as described above with respect to analysis module 312 of
Integrative Approach to Detecting Security Threats
As discussed above, conventional email filtering services are not suitable for recognizing sophisticated malicious emails, and therefore may allow sophisticated malicious emails to reach employees' inboxes by mistake. Introduced here are threat detection platforms designed to take an integrative approach to detecting the security threats to an enterprise.
Unlike conventional email filtering services, a threat detection platform can be completely integrated within the enterprise environment. For example, the threat detection platform may receive input indicative of an approval by an individual (e.g., an administrator associated with the enterprise or an administrator of the email service employed by the enterprise) to access email, active directory, mail groups, identity security events, risk events, documents, etc. The approval may be given through an interface generated by the threat detection platform. For example, the individual may access the interface generated by the threat detection platform and then approve access to these resources as part of a registration process.
Upon receiving the input, the threat detection platform can establish a connection with storage medium(s) that include these resources via application programming interface(s) (APIs). For example, the threat detection platform may establish, via an API, a connection with a computer server managed by the enterprise or some other entity on behalf of the enterprise.
The threat detection platform can then download resources from the storage medium(s) to build an ML model that can be used to identify email-based security threats. Thus, the threat detection platform can build an ML model based on retrospective information in order to better identify security threats in real time as emails are received. For example, the threat detection platform may ingest incoming emails and/or outgoing emails corresponding to the last six months, and then the threat detection platform may build an ML model that understands the norms of communication with internal contacts (e.g., other employees) and/or external contacts (e.g., vendors) for the enterprise. Thus, actual threats rather than synthetic threats can be used to identify whether an incoming email is representative of a security threat.
Such an approach allows the threat detection platform to employ an effective ML model nearly immediately upon receiving approval from the enterprise to deploy it. Most standard integration solutions, such as anti-spam filters, will only have access going forward in time (i.e., after receiving the approval). Here, however, the threat detection platform may employ a backward-looking approach to develop personalized ML model(s) that are effective immediately. Moreover, such an approach enables the threat detection platform to go through a repository of past emails to identify security threats residing in employees' inboxes.
The aforementioned API-based approach provides a consistent, standard way of looking at all email handled by an enterprise (or another entity, such as an email service, on behalf of the enterprise). This includes internal-to-internal email that is invisible from standard integration solutions. A SEG integration, for example, that occurs through the mail exchanger (MX) record will only be able to see incoming email arriving from an external source. The only way to make email arriving from an internal source visible to the SEG integration would be to externally reroute the email through the gateway.
The threat detection platform can design the ML model to sort emails determined to be possible security threats into multiple categories.
Then, the threat detection platform will perform an entity resolution procedure in order to identify the entities involved in the event. Generally, the entity resolution procedure is a multi-step process.
First, the threat detection platform will acquire information regarding the event. For example, if the event is the receipt of an incoming email, the threat detection platform may examine the incoming email to identify the origin, sender identity, sender email address, recipient identity, recipient email address, subject, header(s), body content, etc. Moreover, the threat detection platform may be able to determine whether the incoming email includes any links, attachments, etc.
Second, the threat detection platform will resolve the entities involved in the event by examining the acquired information.
In some embodiments, the threat detection platform will augment the acquired information with human-curated content. For example, feature(s) of an entity may be extracted from human-curated datasets of well-known brands, domains, etc. These human-curated datasets may be used to augment information gleaned from the enterprise's own datasets. Additionally or alternatively, humans may be responsible for labeling entities in some situations. For example, a human may be responsible for labeling landing pages and/or Uniform Resource Locators (URLs) of links found in incoming emails. Human involvement may be useful when quality control is a priority, when comprehensive labeling of evaluation metrics is desired, etc. For example, a human may actively select which data/entities should be used for training the ML model(s) used by the threat detection platform.
The threat detection platform can then establish, compile, and/or calculate corpus statistics for the entities determined to be involved in the event. These corpus statistics can be stored/visualized in terms of an entity risk graph. As shown in
During a second stage, the threat detection platform can extract text, metadata, and/or signals (collectively referred to as “extracted items”) from the downloaded data. For example, the threat detection platform may identify an attachment signal in an email using learned model parameter(s) for text extraction. The term “extracted signal,” as used herein, refers to any information, raw or derived, used by the algorithm(s) employed by the threat detection platform as input. Examples of extracted signals include, but are not limited to, structured data such as IP addresses, third-party data or datasets, API-based integration information with any third-party tool, or other enterprise data or datasets. The extracted items can be persisted in a columnar format in which each column is independently updated. As shown in
During a third stage, the threat detection platform can identify the entities involved in a communication activity. For example, if the communication activity is the receipt of an incoming email, the threat detection platform may identify the sender identity, sender email address, or topic based on the text, metadata, and/or signals extracted during the second stage. As discussed above, humans may be responsible for assisting in entity resolution in some instances. Thus, the third stage may be partially performed by human(s) and partially performed by the threat detection platform, or may be entirely performed by the threat detection platform.
During a fourth stage, the threat detection platform can produce summaries for the entities (also referred to as “attributes” of the email) identified in the third stage based on the past communication activities involving those entities. That is, the threat detection platform may produce corpus statistics that are representative of risk scores associated with historical combinations of the entities identified in the third stage. These corpus statistics can be stored/visualized in terms of an entity risk graph, as shown in
During a fifth stage, the threat detection platform can produce a score indicative of risk to the enterprise. Scores may be produced on a per-communication, per-attack-type, or per-entity basis. Accordingly, the threat detection platform may score each incoming email directed to an employee of the enterprise to determine which incoming emails, if any, should be prevented from reaching the employee's inbox. Generally, incoming emails are scored based on compromise score, and compromise is scored based on the number/type of malicious emails that are received. For example, the threat detection platform may include a threat detection engine and an account compromise engine that separately consider incoming emails. The output produced by each engine (e.g., in terms of score, suspiciousness, etc.) may be used by the other engine for better detection. For example, if an account is determined to be in the suspicious range by the account compromise engine, all emails originating from that account may be monitored more sensitively by the threat detection engine. This protects against cases where an unauthorized entity (also referred as an “attacker”) takes over an account and then uses that account to mount phishing attacks. Scoring of communication activities is further discussed below with respect to
In some embodiments, the threat detection platform also “hydrates” the entities identified in the third stage. The term “hydrate,” as used herein, refers to the action of appending additional signals to a communication such as an email. These additional signals can be defined in three locations: (1) code-defined extractor (e.g., secondary attributes); (2) model application (e.g., URL extraction model, ransom model, employee impersonation model); and (3) rule (e.g., specific whitelists or blacklists). As shown in
During a sixth stage, the threat detection platform can compare each score with a threshold to determine how the email should be classified. For example, the threat detection platform may determine whether to classify the email as borderline, suspicious, or bad. The thresholds used to determine how each email should be classified may be controlled using live data and/or replay from logs to determine the acceptable number of flagged messages. In some embodiments, the thresholds are updated, continually or periodically, to maintain a target flag rate. For example, the threat detection platform may alter the threshold so that a predetermined percentage of all incoming emails (e.g., 0.1%, 0.5%, or 1.0%) are flagged as borderline, suspicious, or bad. The threshold for a given model may be calibrated based on an internal target for the number of false positives and/or false negatives generated by the given model. Generally, increasing a threshold will result in the model having fewer false positives at the expense of more false negatives, while lowering a threshold will result in the model having fewer false negatives at the expense of more false positives.
Threat Intelligence
The threat detection platform can be designed to serve as a centralized system that captures indicators of compromise (IOC) gleaned from a variety of sources, including internal sources (e.g., enterprise databases) and external sources. Examples of IOCs include IP addresses, email addresses, URLs, domains, etc. A database of IOCs can be used for a few different purposes. While the most paramount purpose is the detection of incoming emails representative of security threats, the database could be provided to enterprises for ingestion into other security products, such as firewalls, security orchestration, automation and response (SOAR) tools, etc.
At a high level, the threat detection platform can be designed to perform various tasks including:
Embodiments of the threat detection platform may also be designed to permit enabling/disabling of IOCs on a per-enterprise basis. For example, an enterprise may upload a list of IOCs to the threat detection platform that should be used specifically when examining their emails. Moreover, the threat detection platform may annotate IOCs with a probability so that those IOCs which are probably malicious can be supported. Thus, the threat detection platform could be designed to flag those emails determined to be malicious, as well as those emails that may be malicious. In some embodiments, the threat detection platform is able to place a time limit on each IOC to prevent permanent blacklisting. For example, if a given website is found to be hosting a phishing website, the threat detection platform may capture the given website as an IOC for a specified time period after which it checks whether the given website is still hosting the phishing website.
The threat detection platform may overlay the IOCs with attacks discovered, for example, by examining incoming email. That is, the threat detection platform may attempt to match the IOCs with corresponding attacks so that the score calculated for each attack can be attributed to the appropriate IOC. Thereafter, the threat detection platform may filter the IOCs (e.g., based on the scores that have been attributed to them) and then use the filtered IOCs (and corresponding scores) to further bolster the ability to detect security threats.
In some embodiments, the threat detection platform may utilize its ecosystem of multiple enterprises to offer federated capabilities. For example, the threat detection platform could build a central vendor database across its entire environment to establish a list of vendors and learn what constitutes normal behavior for each vendor. For example, the central vendor database may specify the email endpoints used by each vendor, the accountant(s) responsible for sending invoices for each vendor, the invoicing software used by each vendor, the routing/bank account numbers for each vendor, the location from which the invoices of each vendor originate, etc. As another example, the threat detection platform could build a central threat database across its entire environment to establish a list of entities (e.g., IP addresses, URLs, domains, email addresses) that are most notable in sending attacks. The central threat database may be helpful as it permits the threat detection platform to apply knowledge gained from one enterprise across the entire ecosystem. As another example, the threat detection platform may automatically monitor inboxes to which employees have been instructed to forward suspicious email. When the threat detection platform discovers a malicious email that was missed by its ML models, the threat detection platform may automatically pull the malicious email from all other inboxes in the enterprise where it was found. Moreover, the threat detection platform may use its federated ecosystem to pull the malicious email from inboxes of other enterprises.
Generally, the threat detection platform is designed so that datasets can be computed, tracked, and added to the modeling pipeline in which ML models are developed, trained, etc. Each dataset may be readily reproducible, updatable, and searchable/viewable. As noted above, the datasets may be editable through interfaces generated by the threat detection platform. For example, a human may label different elements included in a dataset for the purpose of training an ML model. Examples of databases that may be accessible to the threat detection platform include:
Employee Account Compromise
Then, the threat detection platform employs multiple detectors to score the user compromise signals. Each score may be representative of how highly the user compromise signal corresponds to the likelihood that an employee's account has been compromised. Accordingly, the user compromise signals may be discovered on a per-user basis (e.g., for each employee of an enterprise).
The threat detection platform can detect instances of EAC by comparing user activity to the scored user compromise signals. For example, the threat detection platform may discover, based on the location and/or the frequency of sign ins, that a given user's account may have become compromised. However, the threat detection platform need not necessarily take action immediately. For instance, the threat detection platform may determine what actions, if any, to take based on which user compromise signal(s) indicate abnormal behavior, the score(s) of those user compromise signal(s), etc. As an example, the threat detection platform may take immediate action to prevent further accesses of the account if the relevant user compromise signal(s) have high scores, but the threat detection platform may continue to monitor the account if the relevant user compromise signal(s) have low scores.
Methodologies for Accurate Scoring
The term “accurate scoring” covers a combination of several concepts further discussed above.
First, a threat detection platform may employ one or more ML models, such as deep learning models, to consume the universe of features that have been extracted for an incoming email from primary and secondary attributes to identify potential security threats. Collectively, these ML model(s) may be referred to as the “ML detector.” In some embodiments, a real-time proportional-integral-derivative (PID) controller is used to tune the threshold for each entity whose emails are being monitored to take into consideration the changing landscape of attack type, email content, etc. The thresholds ensure that the ML model(s) have high precision and continue to be highly precise over time. To cover the general attack landscape, the threat management platform may employ a combination of federated ML models and enterprise-specific ML models able to capture the per-enterprise nuances of sophisticated attacks (e.g., spear phishing attacks).
Second, the threat detection platform can glean the signatures of IOCs in real time to determine the nature of any security threats identified by the ML detector (step 2102). Examples of IOCs include IP addresses, email addresses, URLs, domains, etc. For zero-hour attacks, as email-based attacks are identified by the ML detector, the IOCs can be extracted from the emails. These IOCs can be automatically ingested into a database as “signatures” in real time. Thereafter, the signatures can be used in conjunction with the ML detector to discover future attacks with the same features.
Third, the threat detection platform can perform deep feature extraction to identify zero-hour attacks (step 2103). Identifying zero-hour attacks requires deeper content analysis to understand the nuances of a possible attack. For example, deep learning sub-model(s) may be applied to understand the text, content, sentiment, and/or tone of an email. As another example, to find phishing pages, computer vision may be used to compare the landing page of a link embedded in an email to a set of known sign-on pages. As another example, webpage crawling may be performed to extract information regarding a deep link (e.g., a link embedded in an attachment or a link accessible on a linked website) to discover instances of deep phishing.
Threat Detection, Assessment, and Remediation
The threat detection platform can download a first portion of the first data into a local processing environment (step 2002). For example, the threat detection platform may download all emails received by the employee over the last 3 months, 6 months, 12 months, etc. Then, the threat detection platform can build a personalized ML model for the employee based on the first portion of the first data (step 2003). For example, the threat detection platform may parse each email included in the first data to discover one or more attributes, and then the threat detection platform can provide these attributes to the ML model as input for training. Examples of attributes include the sender name, sender email address, subject, etc. Because the personalized ML model is trained using past emails received by the employee, normal communication habits can be established immediately upon deployment.
Thereafter, the threat detection platform may receive second data related to an email received by the employee (step 2004). The threat detection platform can establish whether the email is representative of a security risk by applying the personalized ML model to the second data (step 2005). Such action will cause the personalized ML model to produce an output that indicates whether the email is indicative of an attack. For example, the output may specify whether the email is suspicious, or the output may specify that the email does not conform with the employee's past communication activities.
In some instances, the threat detection platform will determine, based on the output, that the email is indicative of an attack (step 2006). In such instances, the threat detection platform can characterize the attack along multiple dimensions (also referred to as “facets”) (step 2007). For example, the threat detection platform may establish the identity of the attacked party, the attack vector, the identity of the impersonated party, the impersonation strategy, and/or the attack goal.
Other steps may also be included in some embodiments. For example, the threat detection platform may download a second portion of the first data corresponding to the local processing environment. The second portion of the first data may correspond to a different interval of time than the first portion of the first data. For example, the first portion of the first data may include all emails received by the employee over the last 6 months, and the second portion of the first data may include all emails received by the employee from 6-12 months ago. Then, the threat detection platform can establish whether any emails included in the second portion of the first data are representative of security risks by applying the personalized ML model to the second portion of the first data.
In some instances, the threat detection platform will determine, based on the first output, that the email is representative of a malicious email (step 2103). In such instances, the threat detection platform can apply a second model to the email to produce a second output indicative of whether the email is representative of a given type of malicious email (step 2104). As discussed above, the second model may be one or multiple models applied to the email responsive to determining that the email is representative of a malicious email. Thus, the threat detection platform may apply multiple models to the email to produce multiple outputs, and each model of the multiple models may correspond to a different type of malicious email.
The threat detection platform can then determine whether to remediate the email based on the second output (step 2505). That is, the threat detection platform can determine what actions, if any, should be performed based on the second output. For example, if the second output indicates that the email includes a link to an HTML resource, the threat detection platform may follow the link so that the HTML resource is accessed using a virtual web browser, extract a Document Object Model (DOM) for the HTML resource through the virtual web browser, and analyze the DOM to determine whether the link represents a security risk. As another example, if the second output indicates that the email includes a primary link to a resource hosted by a network-accessible hosting service (e.g., Google Drive®, Box®, Dropbox®, or Microsoft OneDrive®), the threat detection platform may follow the primary link so that the resource is accessed using a virtual web browser, discover whether any secondary links to secondary resources are present by examining content of the resource through the virtual web browser, follow each secondary link to analyze content of the corresponding secondary resource through the virtual web browser, and determine whether the primary link represents a security threat based on whether any secondary links were determined to represent security threats. As another example, if the second output indicates that the email includes a link to an HTML resource, the threat detection platform may follow the link so that the HTML resource is accessed using a virtual web browser, capture a screenshot of the HTML resource through the virtual web browser, apply a computer vision (CV) algorithm designed to identify similarities between the screenshot and a library of verified sign-in websites, and determine whether the link represents a security threat based on an output produced by the CV algorithm. As another example, if the second output indicates that the email includes an attachment, the threat detection platform may open the attachment within a secure processing environment and then determine whether the attachment represents a security threat based on an analysis of content of the attachment. For instance, the threat detection platform may use a headless browser instance running on a standalone computer server (also referred to as a “sandbox computer server”) to inspect the attachment (e.g., by generating a screenshot of its contents), rather than open the attachment directly on a computing device associated with the recipient of the email. Moreover, the threat detection platform may examine any links included in the attachment as discussed above.
In some embodiments, the threat detection platform is further configured to apply a third model designed to produce a comprehensible visualization component based on the second output (step 2106). In embodiments where the second model is part of an ensemble of models applied by the threat detection platform, the third model can aggregate the outputs produced by the models in the ensemble, characterize the attack based on the aggregated outputs, and then convert the aggregated outputs into a comprehensible visualization component. For example, the third model may generate a notification that identifies the type of security threat posed by the email, whether remediation actions are necessary, etc. The comprehensible visualization component may be created so that security professionals responsible for addressing/mitigating security threats can more easily understand why an incoming email was flagged as being representative of an attack.
Processing System
The processing system 2200 may include one or more central processing units (“processors”) 2202, main memory 2206, non-volatile memory 2210, network adapter 2212 (e.g., network interface), video display 2218, input/output devices 2220, control device 2222 (e.g., keyboard and pointing devices), drive unit 2224 including a storage medium 2226, and signal generation device 2230 that are communicatively connected to a bus 2216. The bus 2216 is illustrated as an abstraction that represents one or more physical buses and/or point-to-point connections that are connected by appropriate bridges, adapters, or controllers. The bus 2216, therefore, can include a system bus, a Peripheral Component Interconnect (PCI) bus or PCI-Express bus, a HyperTransport or industry standard architecture (ISA) bus, a small computer system interface (SCSI) bus, a universal serial bus (USB), an IIC (I2C) bus, or an Institute of Electrical and Electronics Engineers (IEEE) standard 1394 bus (also referred to as “Firewire”).
The processing system 2200 may share a similar computer processor architecture as that of a desktop computer, tablet computer, personal digital assistant (PDA), mobile phone, game console, music player, wearable electronic device (e.g., a watch or fitness tracker), network-connected (“smart”) device (e.g., a television or home assistant device), virtual/augmented reality systems (e.g., a head-mounted display), or another electronic device capable of executing a set of instructions (sequential or otherwise) that specify action(s) to be taken by the processing system 2200.
While the main memory 2206, non-volatile memory 2210, and storage medium 2226 (also called a “machine-readable medium”) are shown to be single mediums, the term “machine-readable medium” and “storage medium” should be taken to include a single medium or multiple media (e.g., a centralized/distributed database and/or associated caches and servers) that store one or more sets of instructions 2228. The term “machine-readable medium” and “storage medium” shall also be taken to include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the processing system 2200.
In general, the routines executed to implement the embodiments of the disclosure may be implemented as part of an operating system or a specific application, component, program, object, module, or sequence of instructions (collectively referred to as “computer programs”). The computer programs typically comprise one or more instructions (e.g., instructions 2204, 2208, 2228) set at various times in various memory and storage devices in a computing device. When read and executed by the one or more processors 2202, the instruction(s) cause the processing system 2200 to perform operations to execute elements involving the various aspects of the disclosure.
Moreover, while embodiments have been described in the context of fully functioning computing devices, those skilled in the art will appreciate that the various embodiments are capable of being distributed as a program product in a variety of forms. The disclosure applies regardless of the particular type of machine or computer-readable media used to actually affect the distribution.
Further examples of machine-readable storage media, machine-readable media, or computer-readable media include recordable-type media such as volatile and non-volatile memory devices 2210, floppy and other removable disks, hard disk drives, optical disks (e.g., Compact Disk Read-Only Memory (CD-ROMS), Digital Versatile Disks (DVDs)), and transmission-type media such as digital and analog communication links.
The network adapter 2212 enables the processing system 2200 to mediate data in a network 2214 with an entity that is external to the processing system 2200 through any communication protocol supported by the processing system 2200 and the external entity. The network adapter 2212 can include a network adaptor card, a wireless network interface card, a router, an access point, a wireless router, a switch, a multilayer switch, a protocol converter, a gateway, a bridge, bridge router, a hub, a digital media receiver, and/or a repeater.
The network adapter 2212 may include a firewall that governs and/or manages permission to access/proxy data in a computer network, and tracks varying levels of trust between different machines and/or applications. The firewall can be any number of modules having any combination of hardware and/or software components able to enforce a predetermined set of access rights between a particular set of machines and applications, machines and machines, and/or applications and applications (e.g., to regulate the flow of traffic and resource sharing between these entities). The firewall may additionally manage and/or have access to an access control list that details permissions including the access and operation rights of an object by an individual, a machine, and/or an application, and the circumstances under which the permission rights stand.
The techniques introduced here can be implemented by programmable circuitry (e.g., one or more microprocessors), software and/or firmware, special-purpose hardwired (i.e., non-programmable) circuitry, or a combination of such forms. Special-purpose circuitry can be in the form of one or more application-specific integrated circuits (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), etc.
The foregoing description of various embodiments of the claimed subject matter has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed. Many modifications and variations will be apparent to one skilled in the art. Embodiments were chosen and described in order to best describe the principles of the invention and its practical applications, thereby enabling those skilled in the relevant art to understand the claimed subject matter, the various embodiments, and the various modifications that are suited to the particular uses contemplated.
Although the Detailed Description describes certain embodiments and the best mode contemplated, the technology can be practiced in many ways no matter how detailed the Detailed Description appears. Embodiments may vary considerably in their implementation details, while still being encompassed by the specification. Particular terminology used when describing certain features or aspects of various embodiments should not be taken to imply that the terminology is being redefined herein to be restricted to any specific characteristics, features, or aspects of the technology with which that terminology is associated. In general, the terms used in the following claims should not be construed to limit the technology to the specific embodiments disclosed in the specification, unless those terms are explicitly defined herein. Accordingly, the actual scope of the technology encompasses not only the disclosed embodiments, but also all equivalent ways of practicing or implementing the embodiments.
The language used in the specification has been principally selected for readability and instructional purposes. It may not have been selected to delineate or circumscribe the subject matter. It is therefore intended that the scope of the technology be limited not by this Detailed Description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of various embodiments is intended to be illustrative, but not limiting, of the scope of the technology as set forth in the following claims.
This application is a continuation of U.S. patent application Ser. No. 16/672,854, titled THREAT DETECTION PLATFORMS FOR DETECTING, CHARACTERIZING, AND REMEDIATING EMAIL-BASED THREATS IN REAL TIME and filed on Nov. 4, 2019, which claims priority to U.S. Provisional Application No. 62/782,158, titled “Systems and Methods for Email-Based Threat Detection” and filed on Dec. 19, 2018, U.S. Provisional Application No. 62/807,888, titled “Incident Detection and Management” and filed on Feb. 20, 2019, and U.S. Provisional Application No. 62/813,603, titled “Systems and Methods for Email-Based Threat Detection” and filed on Mar. 4, 2019. Each of these applications is incorporated by reference herein its entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5999932 | Paul | Dec 1999 | A |
| 6023723 | McCormick | Feb 2000 | A |
| 6088717 | Reed | Jul 2000 | A |
| 7263506 | Lee | Aug 2007 | B2 |
| 7451487 | Oliver | Nov 2008 | B2 |
| 7610344 | Mehr | Oct 2009 | B2 |
| 7953814 | Chasin | May 2011 | B1 |
| 8566938 | Prakash | Oct 2013 | B1 |
| 8819819 | Johnston | Aug 2014 | B1 |
| 8935788 | Diao | Jan 2015 | B1 |
| 9009824 | Chen | Apr 2015 | B1 |
| 9154514 | Prakash | Oct 2015 | B1 |
| 9213827 | Li | Dec 2015 | B2 |
| 9245115 | Jakobsson | Jan 2016 | B1 |
| 9245225 | Winn | Jan 2016 | B2 |
| 9264418 | Crosley | Feb 2016 | B1 |
| 9348981 | Hearn | May 2016 | B1 |
| 9473437 | Jakobsson | Oct 2016 | B1 |
| 9516053 | Muddu | Dec 2016 | B1 |
| 9537880 | Jones | Jan 2017 | B1 |
| 9571512 | Ray | Feb 2017 | B2 |
| 9686308 | Srivastava | Jun 2017 | B1 |
| 9756007 | Stringhini | Sep 2017 | B1 |
| 9774626 | Himler | Sep 2017 | B1 |
| 9781152 | Mistratov | Oct 2017 | B1 |
| 9847973 | Jakobsson | Dec 2017 | B1 |
| 9946789 | Li | Apr 2018 | B1 |
| 9961096 | Pierce | May 2018 | B1 |
| 9967268 | Hewitt | May 2018 | B1 |
| 10015182 | Shintre | Jul 2018 | B1 |
| 10044745 | Jones | Aug 2018 | B1 |
| 10091312 | Khanwalkar | Oct 2018 | B1 |
| 10104029 | Chambers | Oct 2018 | B1 |
| 10129194 | Jakobsson | Nov 2018 | B1 |
| 10129288 | Xie | Nov 2018 | B1 |
| 10243989 | Ding | Mar 2019 | B1 |
| 10250624 | Mixer | Apr 2019 | B2 |
| 10277628 | Jakobsson | Apr 2019 | B1 |
| 10362057 | Wu | Jul 2019 | B1 |
| 10397272 | Bruss | Aug 2019 | B1 |
| 10419468 | Glatfelter | Sep 2019 | B2 |
| 10523609 | Subramanian | Dec 2019 | B1 |
| 10601865 | Mesdaq | Mar 2020 | B1 |
| 10616272 | Chambers | Apr 2020 | B2 |
| 10673880 | Pratt | Jun 2020 | B1 |
| 10721195 | Jakobsson | Jul 2020 | B2 |
| 10834127 | Yeh | Nov 2020 | B1 |
| 10911489 | Chechik | Feb 2021 | B1 |
| 10972483 | Thomas | Apr 2021 | B2 |
| 10972485 | Ladnai | Apr 2021 | B2 |
| 11019076 | Jakobsson | May 2021 | B1 |
| 11063897 | Kessler | Jul 2021 | B2 |
| 11153351 | Kalinin | Oct 2021 | B2 |
| 20020002520 | Gatto | Jan 2002 | A1 |
| 20020116463 | Hart | Aug 2002 | A1 |
| 20030204569 | Andrews | Oct 2003 | A1 |
| 20040030913 | Liang | Feb 2004 | A1 |
| 20040117450 | Campbell | Jun 2004 | A1 |
| 20040128355 | Chao | Jul 2004 | A1 |
| 20040215977 | Goodman | Oct 2004 | A1 |
| 20040260922 | Goodman | Dec 2004 | A1 |
| 20050039019 | Delany | Feb 2005 | A1 |
| 20050187934 | Motsinger | Aug 2005 | A1 |
| 20050198518 | Kogan | Sep 2005 | A1 |
| 20060036698 | Hebert | Feb 2006 | A1 |
| 20060053203 | Mijatovic | Mar 2006 | A1 |
| 20060191012 | Banzhof | Aug 2006 | A1 |
| 20060253581 | Dixon | Nov 2006 | A1 |
| 20070074169 | Chess | Mar 2007 | A1 |
| 20070276851 | Friedlander | Nov 2007 | A1 |
| 20080005249 | Hart | Jan 2008 | A1 |
| 20080086532 | Cunningham | Apr 2008 | A1 |
| 20080114684 | Foster | May 2008 | A1 |
| 20080201401 | Pugh | Aug 2008 | A1 |
| 20090037350 | Rudat | Feb 2009 | A1 |
| 20090132490 | Okraglik | May 2009 | A1 |
| 20100115040 | Sargent | May 2010 | A1 |
| 20100211641 | Yih | Aug 2010 | A1 |
| 20100318614 | Sager | Dec 2010 | A1 |
| 20110173142 | Dasgupta | Jul 2011 | A1 |
| 20110179126 | Wetherell | Jul 2011 | A1 |
| 20110213869 | Korsunsky | Sep 2011 | A1 |
| 20110214157 | Korsunsky | Sep 2011 | A1 |
| 20110231510 | Korsunsky | Sep 2011 | A1 |
| 20110231564 | Korsunsky | Sep 2011 | A1 |
| 20110238855 | Korsunsky | Sep 2011 | A1 |
| 20120028606 | Bobotek | Feb 2012 | A1 |
| 20120110672 | Judge | May 2012 | A1 |
| 20120137367 | Dupont | May 2012 | A1 |
| 20120233662 | Scott-Cowley | Sep 2012 | A1 |
| 20120278887 | Vitaldevara | Nov 2012 | A1 |
| 20120290712 | Walter | Nov 2012 | A1 |
| 20120297484 | Srivastava | Nov 2012 | A1 |
| 20130041955 | Chasin | Feb 2013 | A1 |
| 20130086180 | Midgen | Apr 2013 | A1 |
| 20130086261 | Lim | Apr 2013 | A1 |
| 20130097709 | Basavapatna | Apr 2013 | A1 |
| 20130167207 | Davis | Jun 2013 | A1 |
| 20130191759 | Bhogal | Jul 2013 | A1 |
| 20140013441 | Hencke | Jan 2014 | A1 |
| 20140032589 | Styler | Jan 2014 | A1 |
| 20140181223 | Homsany | Jun 2014 | A1 |
| 20140325662 | Foster | Oct 2014 | A1 |
| 20140365303 | Vaithilingam | Dec 2014 | A1 |
| 20140379825 | Speier | Dec 2014 | A1 |
| 20140380478 | Canning | Dec 2014 | A1 |
| 20150026027 | Priess | Jan 2015 | A1 |
| 20150128274 | Giokas | May 2015 | A1 |
| 20150143456 | Raleigh | May 2015 | A1 |
| 20150161609 | Christner | Jun 2015 | A1 |
| 20150161611 | Duke | Jun 2015 | A1 |
| 20150228004 | Bednarek | Aug 2015 | A1 |
| 20150234831 | Prasanna Kumar | Aug 2015 | A1 |
| 20150237068 | Sandke | Aug 2015 | A1 |
| 20150295942 | Tao | Oct 2015 | A1 |
| 20150295945 | Canzanese, Jr. | Oct 2015 | A1 |
| 20150319157 | Sherman | Nov 2015 | A1 |
| 20150339477 | Abrams | Nov 2015 | A1 |
| 20160014151 | Prakash | Jan 2016 | A1 |
| 20160036829 | Sadeh-Koniecpol | Feb 2016 | A1 |
| 20160057167 | Bach | Feb 2016 | A1 |
| 20160156654 | Chasin | Jun 2016 | A1 |
| 20160227367 | Alsehly | Aug 2016 | A1 |
| 20160253598 | Yamada | Sep 2016 | A1 |
| 20160262128 | Hailpern | Sep 2016 | A1 |
| 20160301705 | Higbee | Oct 2016 | A1 |
| 20160306812 | McHenry | Oct 2016 | A1 |
| 20160321243 | Walia | Nov 2016 | A1 |
| 20160328526 | Park | Nov 2016 | A1 |
| 20160344770 | Verma | Nov 2016 | A1 |
| 20160380936 | Gunasekara | Dec 2016 | A1 |
| 20170041296 | Ford | Feb 2017 | A1 |
| 20170048273 | Bach | Feb 2017 | A1 |
| 20170098219 | Peram | Apr 2017 | A1 |
| 20170111506 | Strong | Apr 2017 | A1 |
| 20170186112 | Polapala | Jun 2017 | A1 |
| 20170214701 | Hasan | Jul 2017 | A1 |
| 20170222960 | Agarwal | Aug 2017 | A1 |
| 20170223046 | Singh | Aug 2017 | A1 |
| 20170230323 | Jakobsson | Aug 2017 | A1 |
| 20170230403 | Kennedy | Aug 2017 | A1 |
| 20170237754 | Todorovic | Aug 2017 | A1 |
| 20170237776 | Higbee | Aug 2017 | A1 |
| 20170251006 | Larosa | Aug 2017 | A1 |
| 20170289191 | Thioux | Oct 2017 | A1 |
| 20170324767 | Srivastava | Nov 2017 | A1 |
| 20170346853 | Wyatt | Nov 2017 | A1 |
| 20180026926 | Nigam | Jan 2018 | A1 |
| 20180027006 | Zimmermann | Jan 2018 | A1 |
| 20180084003 | Uriel | Mar 2018 | A1 |
| 20180084013 | Dalton | Mar 2018 | A1 |
| 20180091453 | Jakobsson | Mar 2018 | A1 |
| 20180091476 | Jakobsson | Mar 2018 | A1 |
| 20180159808 | Pal | Jun 2018 | A1 |
| 20180189347 | Ghafourifar | Jul 2018 | A1 |
| 20180196942 | Kashyap | Jul 2018 | A1 |
| 20180219888 | Apostolopoulos | Aug 2018 | A1 |
| 20180227324 | Chambers | Aug 2018 | A1 |
| 20180295146 | Kovega | Oct 2018 | A1 |
| 20180324297 | Kent | Nov 2018 | A1 |
| 20180375814 | Hart | Dec 2018 | A1 |
| 20190014143 | Syme | Jan 2019 | A1 |
| 20190026461 | Cidon | Jan 2019 | A1 |
| 20190028509 | Cidon | Jan 2019 | A1 |
| 20190052655 | Benishti | Feb 2019 | A1 |
| 20190065748 | Foster | Feb 2019 | A1 |
| 20190068616 | Woods | Feb 2019 | A1 |
| 20190081983 | Teal | Mar 2019 | A1 |
| 20190087428 | Crudele | Mar 2019 | A1 |
| 20190089711 | Faulkner | Mar 2019 | A1 |
| 20190104154 | Kumar | Apr 2019 | A1 |
| 20190109863 | Traore | Apr 2019 | A1 |
| 20190141183 | Chandrasekaran | May 2019 | A1 |
| 20190166161 | Anand | May 2019 | A1 |
| 20190166162 | Anand | May 2019 | A1 |
| 20190190929 | Thomas | Jun 2019 | A1 |
| 20190190936 | Thomas | Jun 2019 | A1 |
| 20190199745 | Jakobsson | Jun 2019 | A1 |
| 20190205511 | Zhan | Jul 2019 | A1 |
| 20190222606 | Schweighauser | Jul 2019 | A1 |
| 20190238571 | Adir | Aug 2019 | A1 |
| 20190311121 | Martin | Oct 2019 | A1 |
| 20190319905 | Baggett | Oct 2019 | A1 |
| 20190319987 | Levy | Oct 2019 | A1 |
| 20190349400 | Bruss | Nov 2019 | A1 |
| 20190384911 | Caspi | Dec 2019 | A1 |
| 20200007502 | Everton | Jan 2020 | A1 |
| 20200021609 | Kuppanna | Jan 2020 | A1 |
| 20200044851 | Everson | Feb 2020 | A1 |
| 20200053111 | Jakobsson | Feb 2020 | A1 |
| 20200053120 | Wilcox | Feb 2020 | A1 |
| 20200068031 | Kursun | Feb 2020 | A1 |
| 20200074078 | Saxe | Mar 2020 | A1 |
| 20200076825 | Vallur | Mar 2020 | A1 |
| 20200125725 | Petersen | Apr 2020 | A1 |
| 20200127962 | Bura | Apr 2020 | A1 |
| 20200162483 | Farhady | May 2020 | A1 |
| 20200204572 | Jeyakumar | Jun 2020 | A1 |
| 20200287936 | Nguyen | Sep 2020 | A1 |
| 20200344251 | Jeyakumar | Oct 2020 | A1 |
| 20200358804 | Crabtree | Nov 2020 | A1 |
| 20200374251 | Warshaw | Nov 2020 | A1 |
| 20200389486 | Jeyakumar | Dec 2020 | A1 |
| 20200396258 | Jeyakumar | Dec 2020 | A1 |
| 20200412767 | Crabtree | Dec 2020 | A1 |
| 20210021612 | Higbee | Jan 2021 | A1 |
| 20210058395 | Jakobsson | Feb 2021 | A1 |
| 20210091962 | Finke | Mar 2021 | A1 |
| 20210092154 | Kumar | Mar 2021 | A1 |
| 20210168161 | Dunn | Jun 2021 | A1 |
| 20210240836 | Hazony | Aug 2021 | A1 |
| 20210272066 | Bratman | Sep 2021 | A1 |
| 20210295179 | Eyal Altman | Sep 2021 | A1 |
| 20210329035 | Jeyakumar | Oct 2021 | A1 |
| 20210336983 | Lee | Oct 2021 | A1 |
| 20210360027 | Boyer | Nov 2021 | A1 |
| 20210374679 | Bratman | Dec 2021 | A1 |
| 20210374680 | Bratman | Dec 2021 | A1 |
| Number | Date | Country |
|---|---|---|
| 107315954 | Nov 2017 | CN |
| Entry |
|---|
| Barngrover, Adam, “Vendor Access Management with IGA”, Saviynt Inc. Apr. 24, 2019 (Apr. 24, 2019) Retrieved on Apr. 17, 2021 (Apr. 17, 2021) from <https://saviynt.com/vendor-access-management-with-iga/> entire document, 2 pp. |
| International Search Report and Written Opinion dated Apr. 24, 2020 of PCT/US2019/067279 (14 pages). |
| Miahjajan, Sonal, et al., “Finding HTML Presentation Failures Using Image Comparison Techniques”, ASE' 14, pp. 91-98 (Year: 2014). |
| Mont, Marco Casassa, “Towards accountable management of identity and privacy: Sticky policies and enforceable tracing services”, 14th International Workshop on Database and Expert Systems Applications, 2003. Proceedings. IEEE, 2003. Mar. 19, 2003 (Mar. 19, 2003), Retrieved on Apr. 17, 2021 (Apr. 17, 2021) from <https://eeexplore.ieee.org/abstract/documenV1232051 > entire document, Mar. 19, 2003, 17 pp. |
| Proofpoint (Proofpoint Closed-Loop Email Analysis and Response, Aug. 2018, 2 pages) (Year: 2018). |
| Information Security Media Group, “Multi-Channel Fraud: A Defense Plan”, Retrieved on Apr. 18, 2021 (Apr. 18, 2021) from <https ://www .bankInfosecurity.com/Interviews/multi-channel-fraud-defense-plan-i-1799>, Feb. 20, 2013, 9 pages. |
| Number | Date | Country | |
|---|---|---|---|
| 20220030018 A1 | Jan 2022 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62813603 | Mar 2019 | US | |
| 62807888 | Feb 2019 | US | |
| 62782158 | Dec 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16672854 | Nov 2019 | US |
| Child | 17498273 | US |
