Systems and methods for selecting client backup files for maliciousness analysis
Disclosed herein are systems and methods for selecting files for malware analysis. In one aspect, a method may include identifying, in a cloud network, a backup of a client machine; extracting, from the backup, at least one file of a given file type; determining whether to include the at least one file in a sandbox of the cloud network by performing a static analysis of the at least one file; selecting the at least one file for inclusion in the sandbox based on the static analysis; monitoring, for a period of time, a behavior of the at least one file in the sandbox by performing a dynamic analysis of the at least one file; and in response to determining that the at least one file is malicious based on the dynamic analysis, performing a remediation action on the at least one file.
1 . A method for selecting files for malware analysis, the method comprising:
identifying, in a cloud network, a backup of a client machine;
extracting, from the backup, at least one file of a given file type;
determining whether to include the at least one file in a sandbox of the cloud network by performing a static analysis of the at least one file, wherein the static analysis comprises determining a likelihood of the at least one file being malicious and comparing the likelihood to a threshold likelihood;
in response to determining that the likelihood exceeds the threshold likelihood, selecting the at least one file for inclusion in the sandbox, wherein the sandbox is a software environment that isolates the at least one file from other files in the backup;
monitoring, for a period of time, a behavior of the at least one file in the sandbox by performing a dynamic analysis of the at least one file, wherein the dynamic analysis comprises classifying a given file as malicious or non-malicious;
when determining that the at least one file is malicious based on the dynamic analysis, performing a remediation action on the at least one file; and
when determining that the at least one file is not classified as malicious over the period of time based on the dynamic analysis, removing the at least one file from the sandbox.
2 . The method of claim 1 , wherein the remediation action comprises one or more of:
removing the at least one file from the backup;
quarantining the at least one file in the backup;
removing the at least one file from the client machine; and
quarantining the at least one file in the client machine and triggering a threat investigation process.
3 . The method of claim 1 , wherein the given file type is one of a script and an executable, wherein the threshold likelihood is different for each file type, and wherein the given file type is determined using one or more of magic numbers, file name extensions, rules, heuristics, and inferences of a trained machine learning algorithm.
4 . The method of claim 1 , wherein determining the likelihood of the at least one file being malicious comprises:
determining whether one or more rules of a plurality of rules indicate that the at least one file is malicious; and
calculating the likelihood based on a weight of the one or more rules.
5 . The method of claim 4 , wherein the plurality of rules query features present in the at least one file and classify whether the features are associated with potentially-malicious files, wherein the features include whether the at least one file was downloaded from a blacklisted website, whether the at least one file has been modified a threshold number of times in a small period of time, whether the at least one file has changed a system setting, and/or whether a hash value of the at least one file matches a known malware hash.
6 . The method of claim 1 , wherein determining the likelihood of the at least one file being malicious comprises:
executing a first machine learning algorithm configured to output a classification of whether the at least one file is to be included in the sandbox and a confidence score of the classification, wherein the confidence score is the likelihood, wherein the first machine learning algorithm is trained to make the classification based on features including operation codes (opcodes), byte sequences, Portable Executable (PE) header, and file size of input files.
7 . The method of claim 1 , wherein classifying the given file as malicious or non-malicious comprises executing a malware scanner that compares the at least one file to virus definitions.
8 . The method of claim 1 , wherein classifying the given file as malicious or non-malicious comprises executing a second machine learning algorithm configured to classify the at least one file as malicious or non-malicious, and wherein the second machine learning algorithm is trained based on a training dataset comprising features of labelled files in a plurality of sandboxes.
9 . The method of claim 1 , wherein in response to determining that the at least one file is not classified as malicious over the period of time based on the dynamic analysis, dissolving the sandbox.
10 . The method of claim 1 , further comprising generating a different sandbox for each backup in a backup archive, wherein the different sandbox is dissolved after a predetermined period of time.
11 . The method of claim 1 , wherein the threshold likelihood is adjusted based on an amount of resources in the cloud network.
12 . The method of claim 1 , wherein the sandbox includes files originating from backups of a plurality of client machines.
13 . A system for selecting files for malware analysis, comprising:
a memory storing instructions; and
a hardware processor communicatively coupled with the memory and configured to execute the instructions to:
identify, in a cloud network, a backup of a client machine;
extract, from the backup, at least one file of a given file type;
determine whether to include the at least one file in a sandbox of the cloud network by performing a static analysis of the at least one file, wherein the static analysis comprises determining a likelihood of the at least one file being malicious and comparing the likelihood to a threshold likelihood;
in response to determining that the likelihood exceeds the threshold likelihood, select the at least one file for inclusion in the sandbox, wherein the sandbox is a software environment that isolates the at least one file from other files in the backup;
monitor, for a period of time, a behavior of the at least one file in the sandbox by performing a dynamic analysis of the at least one file, wherein the dynamic analysis comprises classifying a given file as malicious or non-malicious;
when determining that the at least one file is malicious based on the dynamic analysis, perform a remediation action on the at least one file; and
when determining that the at least one file is not classified as malicious over the period of time based on the dynamic analysis, remove the at least one file from the sandbox.
14 . The system of claim 13 , wherein the remediation action comprises one or more of:
removing the at least one file from the backup;
quarantining the at least one file in the backup;
removing the at least one file from the client machine; and
quarantining the at least one file in the client machine and triggering a threat investigation.
15 . The system of claim 13 , wherein the given file type is one of a script and an executable, wherein the threshold likelihood is different for each file type, and wherein the given file type is determined using one or more of magic numbers, file name extensions, rules, heuristics, and inferences of a trained machine learning algorithm.
16 . The system of claim 13 , wherein the hardware processor is configured to determine the likelihood of the at least one file being malicious by:
determining whether one or more rules of a plurality of rules indicate that the at least one file is malicious; and
calculating the likelihood based on a weight of the one or more rules.
17 . The system of claim 16 , wherein the plurality of rules query features present in the at least one file and classify whether the features are associated with potentially-malicious files, wherein the features include whether the at least one file was downloaded from a blacklisted website, whether the at least one file has been modified a threshold number of times in a small period of time, whether the at least one file has changed a system setting, and/or whether a hash value of the at least one file matches a known malware hash.
18 . The system of claim 13 , wherein the hardware processor is configured to determine the likelihood of the at least one file being malicious by:
executing a first machine learning algorithm configured to output a classification of whether the at least one file is to be included in the sandbox and a confidence score of the classification, wherein the confidence score is the likelihood, wherein the first machine learning algorithm is trained to make the classification based on features including operation codes (opcodes), byte sequences, Portable Executable (PE) header, and file size of input files.
19 . A non-transitory computer readable medium storing thereon computer executable instructions for selecting files for malware analysis, the instructions, when executed comprise:
identifying, in a cloud network, a backup of a client machine;
extracting, from the backup, at least one file of a given file type;
determining whether to include the at least one file in a sandbox of the cloud network by performing a static analysis of the at least one file, wherein the static analysis comprises determining a likelihood of the at least one file being malicious and comparing the likelihood to a threshold likelihood;
in response to determining that the likelihood exceeds the threshold likelihood, selecting the at least one file for inclusion in the sandbox, wherein the sandbox is a software environment that isolates the at least one file from other files in the backup;
monitoring, for a period of time, a behavior of the at least one file in the sandbox by performing a dynamic analysis of the at least one file, wherein the dynamic analysis comprises classifying a given file as malicious or non-malicious;
when determining that the at least one file is malicious based on the dynamic analysis, performing a remediation action on the at least one file; and
when determining that the at least one file is not classified as malicious over the period of time based on the dynamic analysis, removing the at least one file from the sandbox.