Systems and methods for generating cyberattack predictions and responses
Systems and methods are provided for making predictions relating to the attack sequence of an attacker or other malicious entity.
1. A method for generating a predictive response to a cyberattack comprising:
detecting information indicative of a cyberattack on an endpoint;
configuring the information as an input to a directed graph model, wherein the directed graph model comprises a plurality of nodes, each node representing an attack state;
providing the input to the directed graph model;
computing, via the directed graph model, a prediction result, the prediction result comprising one or more predicted nodes from the plurality of nodes and one or more associated probabilities;
classifying the one or more predicted nodes; and
determining, based on the classifying and the prediction result, one or more actions for responding to the cyberattack, the one or more actions comprising:
identifying a plurality of parent nodes of the prediction result; and
at least one of:
setting the one or more actions based on risk levels of the plurality of parent nodes;
if each of the plurality of parent nodes comprise equal risk levels, setting the one or more actions based on a union of actions from the plurality of parent nodes;
setting the one or more actions based on a level of specificity of a parent node;
setting the one or more actions based on probabilities of the plurality of parent nodes; or
if each of the plurality of parent nodes comprise equal probability levels, setting the one or more actions based on a union of actions from the plurality of parent nodes.
2. The method of claim 1 comprising, prior to the determining, filtering the one or more predicted nodes based on one or more of probability, distance from the input, classification, and total number of nodes.
3. The method of claim 2 , wherein filtering the one or more predicted nodes based on probability comprises excluding predicted nodes with a probability less than a pre-defined probability threshold.
4. The method of claim 2 , wherein filtering the one or more predicted nodes based on distance from the input comprises excluding predicted nodes with a distance equal to or above a pre-defined threshold.
5. The method of claim 2 , wherein filtering the one or more predicted nodes based on classification comprises excluding predicted nodes that are gap nodes.
6. The method of claim 1 comprising, prior to the determining, verifying that a list of response actions associated with the one or more predicted nodes is unique.
7. The method of claim 1 , wherein the information comprises a tactic, a technique, a sub-technique, and a software class.
8. The method of claim 1 , wherein configuring the information as an input to a directed graph model comprises:
identifying a node of the plurality of nodes that matches the information; and
demarcating the identified node as evidence.
9. The method of claim 8 , wherein identifying the node that matches the information comprises:
calculating a distance; and
determining that the distance is within a predefined range.
10. The method of claim 9 , wherein the distance is either a Hamming distance or an Earth movers distance.
11. The method of claim 1 , wherein the prediction result further comprises a list of response actions.
12. The method of claim 11 , wherein the list of response actions comprises at least one of isolating a host, resetting a password, disabling a user account, blocking a hash, searching emails, blocking a sender, blocking a URL, deleting an email, blocking network traffic, or a kill process.
13. The method of claim 1 , wherein classifying the one or more predicted nodes comprises determining whether each node is a future prediction, a past prediction, or a gap prediction.
14. The method of claim 1 comprising using the prediction result to re-train or update the directed graph model.
15. The method of claim 1 , wherein determining, based on the classifying and the prediction result, one or more actions for responding to the cyberattack comprises:
identifying a parent node that is defined as a cause of the prediction result; and
setting the one or more actions based on the parent node.
16. The method of claim 1 , wherein configuring the information as an input to a directed graph model comprises defining at least one evidence node.
17. A method for generating a predictive response to a cyberattack comprising:
receiving information indicative of a cyberattack on an endpoint from a third-party service;
configuring the information as an input to a directed graph model, wherein the directed graph model comprises a plurality of nodes, each node representing an attack state;
providing the input to the directed graph model;
computing, via the directed graph model, a prediction result, the prediction result comprising one or more predicted nodes from the plurality of nodes and one or more associated probabilities;
classifying the one or more predicted nodes; and
determining, based on the classifying and the prediction result, one or more actions for responding to the cyberattack, the one or more actions comprising:
identifying a plurality of parent nodes of the prediction result; and
at least one of:
setting the one or more actions based on risk levels of the plurality of parent nodes;
if each of the plurality of parent nodes comprise equal risk levels, setting the one or more actions based on a union of actions from the plurality of parent nodes;
setting the one or more actions based on a level of specificity of a parent node;
setting the one or more actions based on probabilities of the plurality of parent nodes; or
if each of the plurality of parent nodes comprise equal probability levels, setting the one or more actions based on a union of actions from the plurality of parent nodes.
18. The method of claim 17 comprising, prior to the determining, filtering the one or more predicted nodes based on one or more of probability, distance from the input, classification, and total number of nodes.
19. The method of claim 18 , wherein filtering the one or more predicted nodes based on probability comprises excluding predicted nodes with a probability less than a pre-defined probability threshold.
20. The method of claim 18 , wherein filtering the one or more predicted nodes based on distance from the input comprises excluding predicted nodes with a distance equal to or above a pre-defined threshold.
21. The method of claim 18 , wherein filtering the one or more predicted nodes based on classification comprises excluding predicted nodes that are gap nodes.
22. The method of claim 18 comprising, prior to the determining, verifying that a list of response actions associated with the one or more predicted nodes is unique.
23. The method of claim 18 , wherein the information comprises a tactic, a technique, a sub-technique, and a software class.
24. The method of claim 18 , wherein configuring the information as an input to a directed graph model comprises:
identifying a node of the plurality of nodes that matches the information; and
demarcating the identified node as evidence.
25. The method of claim 24 , wherein identifying the node that matches the information comprises:
calculating a distance; and
determining that the distance is within a predefined range.
26. The method of claim 25 , wherein the distance is either a Hamming distance or an Earth movers distance.
27. The method of claim 17 , wherein the prediction result further comprises a list of response actions.
28. The method of claim 27 , wherein the list of response actions comprises at least one of isolating a host, resetting a password, disabling a user account, blocking a hash, searching emails, blocking a sender, blocking a URL, deleting an email, blocking network traffic, or a kill process.
29. The method of claim 17 , wherein classifying the one or more predicted nodes comprises determining whether each node is a future prediction, a past prediction, or a gap prediction.
30. The method of claim 17 , wherein determining, based on the classifying and the prediction result, one or more actions for responding to the cyberattack comprises:
identifying a parent node that is defined as a cause of the prediction result; and
setting the one or more actions based on the parent node.
31. The method of claim 17 , wherein configuring the information as an input to a directed graph model comprises defining at least one evidence node.