Multistage analysis of emails to identify security threats
Conventional email filtering services are not suitable for recognizing sophisticated malicious emails, and therefore may allow sophisticated malicious emails to reach inboxes by mistake. Introduced here are threat detection platforms designed to take an integrative approach to detecting security threats. For example, after receiving input indicative of an approval from an individual to access past email received by employees of an enterprise, a threat detection platform can download past emails to build a machine learning (ML) model that understands the norms of communication with internal contacts (e.g., other employees) and/or external contacts (e.g., vendors). By applying the ML model to incoming email, the threat detection platform can identify security threats in real time in a targeted manner.
1. A system, comprising:
a processor configured to:
receive input indicative of an approval to access emails delivered to an employee of an enterprise;
acquire an incoming email addressed to the employee;
extract a primary attribute from the incoming email by parsing at least one of: (1) content of the incoming email or (2) metadata associated with the incoming email;
determine whether the incoming email deviates from past email activity, at least in part by determining, as a secondary attribute, a mismatch between a previous value for the primary attribute and a current value for the primary attribute, using a communication profile associated with the employee, and feed a measured deviation into at least one machine learning model; and
refine the machine learning model based at least in part on a result of monitoring for a subsequent action taken with respect to the incoming email;
a memory coupled to the processor and configured to provide the processor with instructions.
2. The system of claim 1 , wherein the processor is further configured to establish, via an application programming interface, a connection with an email system employed by the enterprise.
3. The system of claim 1 , wherein the communication profile includes an additional primary attribute and an additional secondary attribute of a past email delivered to the employee determined to be representative of safe communications.
4. The system of claim 3 , wherein determining whether the incoming email deviates from past email activity includes discovering whether the primary attribute, the secondary attribute, or the combination of the primary and secondary attributes is included in the communication profile.
5. The system of claim 1 , wherein the primary attribute is at least one of sender display name, sender username, Sender Policy Framework (SPF) status, DomainKeys Identified Mail (DKIM) status, number of attachments, number of links in a body of the incoming email, country of origin, information in a header of the incoming email, or an identifier embedded in metadata associated with the incoming email.
6. The system of claim 1 , wherein the processor is further configured to:
establish that the incoming email does not represent a security risk; and
update the communication profile by creating an entry that programmatically associates the primary and secondary attributes.
7. The system of claim 6 , wherein said establishing comprises applying a deep learning model to understand content, sentiment, and/or tone of the incoming email.
8. The system of claim 6 , wherein said establishing comprises employing a crawling algorithm to extract information regarding a secondary link that is embedded in an attachment to the incoming email or accessible via a website linked to by a primary link in the incoming email.
9. The system of claim 1 , wherein the processor is further configured to perform an action with respect to the incoming email based at least in part on the determination.
10. The system of claim 9 , wherein performing the action comprises forwarding the incoming email to an inbox of the employee.
11. The system of claim 9 , wherein performing the action further comprises applying a model to the incoming email.
12. The system of claim 11 , wherein the model is one of multiple models applied to the incoming email responsive to determining that the incoming email may be a malicious email.
13. The system of claim 12 , wherein each model of the multiple models is associated with a different type of malicious email.
14. The system of claim 1 , wherein the subsequent action comprises a subsequent action taken by the employee with respect to the incoming email.
15. The system of claim 1 , wherein the subsequent action comprises feedback associated with the incoming email, wherein the feedback is weighted based at least in part on a reputation of an individual providing the feedback.
16. A method, comprising:
receiving input indicative of an approval to access emails delivered to an employee of an enterprise;
acquiring an incoming email addressed to the employee;
extracting a primary attribute from the incoming email by parsing at least one of: (1) content of the incoming email or (2) metadata associated with the incoming email;
determining whether the incoming email deviates from past email activity, at least in part by determining, as a secondary attribute, a mismatch between a previous value for the primary attribute and a current value for the primary attribute, using a communication profile associated with the employee, and feeding a measured deviation into at least one machine learning model; and
refining the machine learning model based at least in part on a result of monitoring for a subsequent action taken with respect to the incoming email.
17. The method of claim 16 , further comprising establishing, via an application programming interface, a connection with an email system employed by the enterprise.
18. The method of claim 16 , wherein the communication profile includes an additional primary attribute and an additional secondary attribute of a past email delivered to the employee determined to be representative of safe communications.
19. The method of claim 18 , wherein determining whether the incoming email deviates from past email activity includes discovering whether the primary attribute, the secondary attribute, or the combination of the primary and secondary attributes is included in the communication profile.
20. The method of claim 16 , wherein the primary attribute is at least one of: sender display name, sender username, Sender Policy Framework (SPF) status, DomainKeys Identified Mail (DKIM) status, number of attachments, number of links in a body of the incoming email, country of origin, information in a header of the incoming email, or an identifier embedded in metadata associated with the incoming email.
21. The method of claim 16 , further comprising:
establishing that the incoming email does not represent a security risk; and
updating the communication profile by creating an entry that programmatically associates the primary and secondary attributes.
22. The method of claim 21 , wherein said establishing comprises applying a deep learning model to understand content, sentiment, and/or tone of the incoming email.
23. The method of claim 21 , wherein said establishing comprises employing a crawling algorithm to extract information regarding a secondary link that is embedded in an attachment to the incoming email or accessible via a website linked to by a primary link in the incoming email.
24. The method of claim 16 , further comprising performing an action with respect to the incoming email based at least in part on the determination.
25. The method of claim 24 , wherein performing the action comprises forwarding the incoming email to an inbox of the employee.
26. The method of claim 24 , wherein performing the action further comprises applying a model to the incoming email.
27. The method of claim 26 , wherein the model is one of multiple models applied to the incoming email responsive to determining that the incoming email may be a malicious email.
28. The method of claim 27 , wherein each model of the multiple models is associated with a different type of malicious email.
29. The method of claim 16 , wherein the subsequent action comprises a subsequent action taken by the employee with respect to the incoming email.
30. The method of claim 16 , wherein the subsequent action comprises feedback associated with the incoming email, wherein the feedback is weighted based at least in part on a reputation of an individual providing the feedback.
31. A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
receiving input indicative of an approval to access emails delivered to an employee of an enterprise;
acquiring an incoming email addressed to the employee;
extracting a primary attribute from the incoming email by parsing at least one of: (1) content of the incoming email or (2) metadata associated with the incoming email;
determining whether the incoming email deviates from past email activity, at least in part by determining, as a secondary attribute, a mismatch between a previous value for the primary attribute and a current value for the primary attribute, using a communication profile associated with the employee, and feeding a measured deviation into at least one machine learning model; and
refining the machine learning model based at least in part on a result of monitoring for a subsequent action taken with respect to the incoming email.