Cyber security using one or more models trained on a normal behavior
Disclosed herein is a method for detection of a cyber-threat to a computer system. The method is arranged to be performed by a processing apparatus. The method comprises receiving input data associated with a first entity associated with the computer system, deriving metrics from the input data, the metrics representative of characteristics of the received input data, analysing the metrics using one or more models, and determining, in accordance with the analysed metrics and a model of normal behavior of the first entity, a cyber-threat risk parameter indicative of a likelihood of a cyber-threat. A computer readable medium, a computer program and a threat detection system are also disclosed.
1. A cyber threat detection system, comprising:
a machine learning model, where at least both of human and machine activity is modelled by the machine learning model configured to ingest data from a number of sources including from an interface for network packet data inspection and an interface for endpoint parameters, where the model is configured to be a self-learning model, where the model is configured to be updated when new data is received,
where the cyber threat detection system is configured to self-learn and detect normality in order to spot true anomalies by understanding a behavior of users and machines, where the cyber threat detection system is configured to output a cyber-threat risk parameter indicative of a cyber-threat, where the cyber-threat includes a human insider threat,
where the model is configured to automate a detection of the human insider threat through malicious behavior of a first user reflected in changes to a pattern of usage in networked office equipment,
where the cyber threat detection system is configured to have at least a processor arranged to run steps of processes and a memory required to store information related to the running of the processes,
and where the outputted cyber-threat risk parameter is determined based at least in part on an absence of expected behavior and a presence of an unexpected behavior.
2. The cyber threat detection system of claim 1 , further comprising:
two or more models, where the model is part of the two or more models, where the two or more models include a first model configured to self-learn and detect normality in order to analyse data for detecting a first type of cyber threat being the human insider attack, where the two or more models include a second model configured to self-learn and detect normality in order to analyse data for detecting a second type of threat being a type of cyber threat other than the human insider attack.
3. The cyber threat detection system of claim 1 , where the cyber threat detection system is configured to protect users and machine, where the cyber threat detection system is configured to build and maintain a dynamic, ever-changing model of a normal behavior of each user and machine protected by the cyber threat detection system.
4. The cyber threat detection system of claim 1 , where the cyber threat detection system is configured to analyze links between data associated with a first entity and data associated with a second entity, where the analyzed links between the entities on a system are taken into consideration when performing a threat detection determination.
5. The cyber threat detection system of claim 1 , where the machine learning model is configured to be a normative model, where the machine learning model is configured to develop a pattern of life based on the data gathered regarding the first user to detect shifts in the pattern; and thus, a change in the behavior of the first user.
6. The cyber threat detection system of claim 5 , where the normative model is used as a moving benchmark, allowing the cyber threat detection system to spot changes in the behavior of the first user that seems to fall outside of the normal pattern of life, and the cyber threat detection system is configured to flag this behavior as anomalous, where the normative model is configured to use unsupervised machine learning.
7. The cyber threat detection system of claim 6 , where a presence of the anomalous behavior is indicative of the cyber threat based and factored into the cyber-threat risk parameter, and
where the pattern of life analysis identifies how a human and/or machine behaves over time.
8. The cyber threat detection system of claim 1 , where a second machine learning model is configured to be a normative model, where the machine learning model is configured to convert the data and the new data inputs into the normative model of the machines that are being modeled by the machine learning model.
9. The cyber threat detection system of claim 8 , where the normative model is configured to see patterns in information and activity and to build an understanding of what is normal at any one time, and what is genuinely anomalous, based on a current threat and network environment in order to control a number of false positives.
10. The cyber threat detection system of claim 8 , where the cyber-threat risk parameter is configured to be projected on a graphical user interface that conveys cyber threats across a packet flow and connection topology corresponding to a computing system being protected by the cyber threat detection system.
11. A method for a cyber threat detection system, comprising:
modelling at least both of human and machine activity with a machine learning model by ingesting data from a number of sources including both network packet data inspection and endpoint parameters, where the model is configured to be a self-learning model and configured to be updated when new data is received;
modelling which automates a detection of a human insider threat through malicious behavior of a first user reflected in changes to a pattern of usage in networked office equipment;
updating with the new data to self-learn as well as using a normality of the ingested data in order to spot true anomalies by understanding a behavior of users and machines; and
outputting a cyber-threat risk parameter indicative of a cyber-threat, where the outputted cyber-threat risk parameter is determined based at least in part on at least one of an absence of an expected behavior or a presence of an unexpected behavior, where the cyber-threat includes the human insider threat.
12. The method for the cyber threat detection system of claim 11 , further comprising
using two or more models, where the model is part of the two or more models, where the two or more models include a first model configured to self-learn and detect normality in order to analyse data for detecting a first type of cyber threat being the human insider attack, where the two or more models include a second model configured to self-learn and detect normality in order to analyse data for detecting a second type of threat being a type of cyber threat other than the human insider attack.
13. The method for the cyber threat detection system of claim 11 , where the cyber threat detection system is configured to protect users and machine, and where the cyber threat detection system is further comprising
building and maintaining a dynamic, ever-changing model of a normal behavior of each user and each machine protected by the cyber threat detection system.
14. The method for the cyber threat detection system of claim 11 , further comprising
analyzing links between data associated with a first entity and data associated with a second entity, where the analyzed links between these entities in a system are taken into consideration when performing a threat detection determination.
15. The method for the cyber threat detection system of claim 11 , further comprising
developing a pattern of life, based on the data gathered regarding the first user, with the machine learning model to detect shifts in the pattern; and thus, a change in the behavior of the first user.
16. The method for the cyber threat detection system of claim 15 , further comprising
using the machine learning model modelling the pattern of life of the first user as a moving benchmark, which allows the cyber threat detection system to spot changes in the behavior of the first user that seems to fall outside of the normal pattern of life, and then flagging this behavior as anomalous; and
where the machine learning model modelling the pattern of life uses unsupervised machine learning.
17. The method for the cyber threat detection system of claim 16 , where a presence of the anomalous behavior is indicative of the cyber threat and factored into the cyber-threat risk parameter, and
where the pattern of life analysis identifies how a human and/or machine behaves over time.
18. The method for the cyber threat detection system of claim 11 , further comprising
converting the data and the new data inputs into a normative model the machines that are being modelled.
19. The method for the cyber threat detection system of claim 18 , further comprising
analyzing patterns in information and activity and building an understanding of what is normal at any one time, and what is genuinely anomalous, based on a current threat and network environment in order to control a number of false positives.
20. The method for the cyber threat detection system of claim 18 , further comprising projecting the cyber-threat risk parameter on a graphical user interface that conveys cyber threats across a packet flow and connection topology corresponding to a computing system being protected by the cyber threat detection system.