User-sourced object reputations

Patent metadata

Field	Value
Publication number	US-12425425-B2
Application number	US-202017066629-A
Country	US
Kind code	B2
Filing date	Oct 9, 2020
Priority date	Oct 9, 2020
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

There is disclosed in one example a computing apparatus, including: a hardware platform including a processor circuit and a memory circuit; first means for accessing a machine learning engine; second means for accessing a user interface; and instructions encoded within the memory to instruct the processor to: load into the machine learning engine via the first means an object prevalence model, including an enterprise-specific prevalence model; provide to the machine learning engine an object set from the enterprise; identify an enterprise-novel object from the object set; solicit and receive via the second means user-sourced feedback for the enterprise-novel object; and act according to the user-sourced feedback.

First claim

What is claimed is:

1. A computer-implemented method of providing user-sourced reputation adjustments to an artificial intelligence malware detection engine, comprising:
computing a malware reputation for a first executable software object via the artificial intelligence malware detection engine, wherein the malware reputation accounts for global and local prevalence of the executable software object, wherein the local prevalence comprises prevalence for a first enterprise, and wherein the global prevalence comprises prevalence for a plurality of enterprises different from the first enterprise;
irrespective of the malware reputation, greenlighting the first executable software object based on an input from a human user, wherein greenlighting is an indication that the first executable object is safe or non-malicious and should be allowed on the first enterprise;
analyzing a second executable software object within the first enterprise using the artificial intelligence malware detection engine, wherein the second executable software object comprises a portable executable having structured information consisting of a number of API calls, a number of DLL calls, a number of import functions, a number of export functions, a starting virtual address, virtual size, and language, wherein the second executable software object is novel according to a local prevalence model for the first enterprise and is not identical to the first executable software object;
assigning the second executable software object a malicious or suspicious reputation based on the analyzing
clustering the second executable software object into a cluster of software objects that includes the first executable software object, wherein clustering comprises computing feature vector distances between software objects; and
based on the clustering, greenlighting the second executable software object.

2. The method of claim 1 , further comprising providing to the human user a user interface to present a feed of greenlit objects.

3. The method of claim 2 , wherein the feed of greenlit objects includes similar objects to the greenlit objects.

4. One or more tangible, non-transitory computer-readable storage media having stored thereon executable instructions to:
compute, from a machine learning engine comprising an object prevalence model, a malware reputation for a first executable software object within a first enterprise;
apply to the first executable software object an adjusted reputation based on user-sourced feedback;
identify a second executable software object within the first enterprise, wherein the second executable software object comprises a portable executable having structured information consisting of a number of API calls, a number of DLL calls, a number of import functions, a number of export functions, a starting virtual address, virtual size, and language, wherein the second executable software object is novel according to a local prevalence model for the first enterprise and is not identical to the first executable software object, wherein the local prevalence comprises prevalence for the first enterprise;
analyze the second executable software object with the machine learning engine and assign to the second executable software object a malicious or suspicious reputation according to the machine learning engine;
cluster the second executable software object into a cluster of software objects that includes the first executable software object, wherein clustering comprises computing feature vector distances between software objects; and
based on the clustering, greenlight the second executable software object.

5. The one or more tangible, non-transitory computer-readable storage media of claim 4 , wherein the instructions are further to provide a guest infrastructure to execute at least some of the instructions.

6. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein the guest infrastructure comprises a virtualization infrastructure.

7. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein the guest infrastructure comprises a containerization infrastructure.

8. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein the object prevalence model further comprises a global prevalence model.

9. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein the user-sourced feedback comprises an assigned reputation for the executable software object.

10. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein clustering comprising computing a distance between the second executable software object and a next-nearest neighbor.

11. The one or more tangible, non-transitory computer-readable storage media of claim 10 , wherein the next-nearest neighbor is an object previously known to the enterprise.

12. The one or more tangible, non-transitory computer-readable storage media of claim 10 , wherein the next-nearest neighbor is a previous version of the second executable software object.

13. The one or more tangible, non-transitory computer-readable storage media of claim 5 , wherein the instructions are further to retrain the machine learning engine with the user-sourced feedback.

14. A computing apparatus, comprising:
a processor circuit;
a memory; and
instructions encoded within the memory to instruct the processor circuit to:
compute, within a machine learning engine comprising an object prevalence model, a malware reputation for a first executable software object, wherein the malware reputation accounts for both local and global prevalence of the object, wherein the local prevalence comprises prevalence for a first enterprise, and wherein the global prevalence comprises prevalence for a plurality of enterprises different from the first enterprise, and wherein computing the malware reputation comprises analyzing structured information consisting of a number of API calls, a number of DLL calls, a number of import functions, a number of export functions, a starting virtual address, virtual size, and language;
apply to the first executable software object an adjusted reputation based on user-sourced feedback;
identify a second executable software object within the first enterprise, wherein the second executable software object is novel according to the local prevalence for the first enterprise and is not identical to the executable software object;
compute, within the machine-learning engine, a malicious or suspicious reputation for the second executable software object;
cluster the second executable software object into a cluster of software objects that includes the first executable software object, wherein clustering comprises computing feature vector distances between software objects; and
based on the clustering, greenlight the second executable software object.

15. The computing apparatus of claim 14 , wherein the instructions are further to provide a live backend system.

16. The computing apparatus of claim 15 , wherein the live backend system is to measure model accuracy over a plurality of segments.

17. The computing apparatus of claim 16 , wherein the plurality of segments includes a segment selected from geographic, industry sector, operating system type, and traffic type.

18. The computing apparatus of claim 14 , wherein the object prevalence model further comprises a global prevalence model.

19. The computing apparatus of claim 14 , wherein the user-sourced feedback

Assignees

• Mcafee Llc

Inventors

• Healy Sorcha Bairbre
• Murphy Gerard Donal
• Grobman Steven
• Fitzgerald Niall
• Daly Jillian Anne
• Thakur Sandeep
• Gaither Brian
• Minihane Niamh
• Costigan Catherine

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title