Method and system for detecting malicious activity
A method and a system for detecting a malicious activity are provided. The method comprises: receiving, from a given host of the plurality of hosts, an event flow including data representative of events occurred at the given host; analyzing a given event sequence of the event flow to generate, for a given event thereof, a respective internal event; applying to the respective internal event, a plurality of signature-based rules to determine at least one internal state marker of the given host associated with the given event; feeding the respective internal state markers to a trained machine-learning algorithm (MLA) to determine a prediction outcome thereof of whether the given event sequence is associated with the malicious activity; in response to the prediction outcome exceeding a predetermined threshold, determining the given event sequence as being associated with the malicious activity; and generating a report including the prediction outcome.
1. A computer-implemented method for detection of a malicious activity by analyzing object behavior in a non-isolated environment, the method comprising:
receiving, from a given host of a plurality of hosts of the non-isolated environment, an event flow including data representative of events that occurred at the given host;
analyzing, a given event sequence of the event flow having been generated for a predetermined period, to generate, for a given event of the given event sequence, a respective internal event, the respective internal event being format-invariant to other events in the event flow;
applying, to the respective internal event, a plurality of signature-based rules to determine at least one internal state marker of the given host associated with the given event of the given event sequence,
the respective internal state marker being indicative of a then current state of the given host during an occurrence of the given event of the given event sequence; and
respective internal state markers associated with the given event sequence being indicative of whether the given event sequence is associated with the malicious activity or not;
feeding the respective internal state markers to a trained machine-learning algorithm (MLA) to determine a prediction outcome thereof of whether the given event sequence is associated with the malicious activity,
the trained MLA having been trained to determine whether the given event sequence is associated with the malicious activity using a training set of data comprising: (i) arrays of training internal state markers associated with training event sequences; and (ii) a respective label assigned to each of the training event sequences, the respective label for a given training event sequence being indicative of whether the given training event sequence is associated with the malicious activity or not;
in response to the prediction outcome exceeding a predetermined threshold value, determining that the given event sequence having occurred at the given host is associated with the malicious activity; and
generating a report including the prediction outcome for presentation thereof to a user.
2. The method of claim 1 , prior to applying the plurality of signature-based rules, further comprising storing the respective internal event in a pipeline of internal events.
3. The method of claim 1 , wherein the determining the at least one internal state marker comprises detecting one of malicious activity and object in one of the given host of the plurality of hosts and a network thereof.
4. The method of claim 3 , wherein the at least one internal state marker is valid for the predetermined period.
5. The method of claim 3 , wherein the at least one internal state marker has been pre-assigned with a respective weight value.
6. The method of claim 1 , wherein at least one internal state indicated by the at least one internal state marker comprises at least one of: a set of flags indicative of the then current state of the given host, file paths, a registry, processes, signatures, mutexes, synchronization objects, and counters.
7. The method of claim 6 , wherein the at least one internal state is valid for the predetermined period.
8. The method of claim 1 , wherein the trained MLA comprises at least one of: a linear model, a classifier, and a neural network.
9. The method of claim 1 , wherein the report further comprises at least one of: signature markers, information of the given host, contents of environment variables of processes running during occurrences of the given event sequence, information about connected devices, a screenshot, a screen video at a moment of the determining the given event sequence as being associated with the malicious activity, a sequence of user actions executed prior to the given event sequence, an incident graph.
10. A system for detecting a malicious activity by analyzing objects behavior in a non-isolated environment, the system comprising: at least one processor communicatively coupled to a plurality of hosts of the non-isolated environment and at least one non-transitory computer-readable memory storing instructions, which when executed by the at least one processor, cause the system to:
receive, from a given host of the plurality of hosts, an event flow including data representative of events that occurred at the given host;
analyze a given event sequence of the event flow having been generated for a predetermined period, to generate, for a given event of the given event sequence, a respective internal event, the respective internal event being format-invariant to other events in the event flow;
apply, to the respective internal event, a plurality of signature-based rules to determine at least one internal state marker of the given host associated with the given event sequence,
the respective internal state marker being indicative of a then current state of the given host during an occurrence of the given event of the given event sequence; and
respective internal state markers associated with the given event sequence being indicative of whether the given event sequence is associated with the malicious activity or not;
feed the respective internal state markers to a trained machine-learning algorithm (MLA) to determine a prediction outcome thereof of whether the given event sequence is associated with the malicious activity,
the trained MLA having been trained to determine whether the given event sequence is associated with the malicious activity using a training set of data comprising: (i) arrays of training internal state markers associated with training event sequences; and (ii) a respective label assigned to each of the training event sequences, the respective label for a given training event sequence being indicative of whether the given training event sequence is associated with the malicious activity or not;
in response to the prediction outcome exceeding a predetermined threshold value, determine that the given event sequence having occurred at the given host is associated with the malicious activity; and
generate a report including the prediction outcome for presentation thereof to a user.