Granted Patent B2

US 12,568,094 · App. 17/864,551 · Granted Mar 3, 2026

Inventor: Anton Victorovich Afonin (Moscow, RU)

Assignee: GROUP-IB GLOBAL PRIVATE LIMITED

H04L63/1416H04L61/4511H04L63/0236H04L63/1466

Abstract

Claims (59)

1 . A computer-implementable method of identifying, in a network infrastructure, compromised network devices that are using DNS tunneling, the computer-implementable method being executable by a computing device located within the network infrastructure, the computing device including at least one processor, the computer-implementable method comprising:

receiving, by the at least one processor, traffic generated by each network device of a plurality of network devices located within the network infrastructure;

identifying, by the at least one processor, from the traffic, a plurality of DNS queries having been generated by respective network devices of the plurality of network devices located within the network infrastructure,

each one of the plurality of DNS queries having been transmitted outside the network infrastructure;

determining, by the at least one processor, in the plurality of DNS queries, presence of DNS queries to predetermined types of domain names, generation of which, by a respective network device of the network infrastructure, is non-indicative of the respective network device being compromised,

the predetermined types of domain names including at least low entropy domain names;

removing, by the at least one processor, from the plurality of DNS queries, the DNS queries to the predetermined types of domain names, thereby generating a refined plurality of DNS queries;

generating, by the at least one processor, for a given one of the refined plurality of DNS queries, a respective set of suspiciousness features, the respective set of suspiciousness features comprising:

statistics of network interactions of the respective network device having transmitted the given one of the refined plurality of DNS queries,

the statistics of network interactions including: (i) for protocols above a Transmission Control Protocol (TCP), a number of TCP sessions; and (ii) for User Datagram Protocol (UDP), a number of datagrams over a predetermined period after the respective network device most recently sent the given one of the refined plurality of DNS queries;

applying, by the at least one processor, to the respective set of suspiciousness features associated with the given one of the refined plurality of DNS queries, a pre-trained decision rule,

the pre-trained decision rule having been trained to determine, based on the respective set of suspiciousness features, whether a respective domain name associated with the given one of the refined plurality of DNS queries is malicious;

in response to the pre-trained decision rule rendering a positive outcome, increasing a respective penalty score for each network device of the network infrastructure having transmitted the given one of the refined plurality of DNS queries;

in response to the respective penalty score associated with a given network device exceeding a predetermined penalty score threshold, identifying, by the at least one processor, the given network device as being compromised; and

generating, by the at least one processor, a warning notification for transmission thereof about the given network device being compromised.

2 . The computer-implementable method of claim 1 , wherein the traffic of the network infrastructure comprises at least one of inbound traffic and outbound traffic.

3 . The computer-implementable method of claim 1 , wherein the receiving the traffic comprises applying, by the at least one processor, a traffic mirroring approach.

4 . The computer-implementable method of claim 1 , wherein the determining, in the plurality of DNS queries, the presence of the DNS queries to the predetermined types of domain names comprises determining, for each one of the plurality of DNS queries, at least one of: a domain name; a sender IP address, and a recipient IP address.

5 . The computer-implementable method of claim 4 , further comprising determining whether at least one recipient IP address is in one of a blacklist and in a whitelist, and

in response to the at least one recipient IP address being in the blacklist, increasing the penalty score for each network device of the network infrastructure having transmitted a respective DNS query including the at least one recipient IP address;

in response to at least one recipient IP address being in the whitelist, identifying the respective DNS query as being one of the DNS queries to the predetermined types of domain names.

6 . The computer-implementable method of claim 1 , wherein the predetermined types of domain names further include:

second level domain names and

domain names having been identified as trusted domain names.

7 . The computer-implementable method of claim 1 , wherein the pre-trained decision rule comprises at least one of (i) a heuristic rule and (ii) a long short-term memory (LSTM) neural network-based rule,

the LSTM neural network-based rule having been determined by a LSTM neural network trained to determine, based on the respective set of suspiciousness features, whether the respective domain name associated with the given one of the refined plurality of DNS queries is malicious.

8 . The computer-implementable method of claim 7 , wherein applying the heuristic rule comprises:

determining, by the at least one processor, for the respective domain name of the given one of the refined plurality of DNS queries, a respective entropy value; and

in response to the respective entropy value being greater than a predetermined entropy value threshold, determining that the respective domain name associated with the given one of the refined plurality of DNS queries is malicious.

9 . The computer-implementable method of claim 7 , wherein applying the heuristic rule comprises:

determining, by the at least one processor, a respective frequency of occurrence, in the respective domain name, of each one of a plurality of predetermined N-grams;

based on respective frequency values, determining an N-gram entropy value for the respective domain name; and

in response to the N-gram entropy value exceeding a predetermined N-gram entropy threshold value, determining that the respective domain name associated with the given one of the refined plurality of DNS queries is malicious.

10 . The computer-implementable method of claim 7 , wherein applying the LSTM neural network-based rule comprises feeding, by the at least one processor, to the LSTM neural network: (i) the respective set of suspiciousness features associated with the given one of the refined plurality of DNS queries; (ii) the statistics of network interactions of the respective network device having transmitted the given one of the refined plurality of DNS queries; and (iii) a current penalty score of the respective network device.

11 . The computer-implementable method of claim 1 , wherein, in response to the pre-trained decision rule rendering the positive outcome, the method further comprises:

determining, by the at least one processor, in past traffic of the network infrastructure, whether there is at least one other network device having transmitted a DNS query including the respective domain name of the given one of the refined plurality of DNS queries;

in response to determining the at least one other network device, executing one of:

increasing the respective penalty score of the at least one other network device; and

increasing respective penalty scores of each network device of the network infrastructure.

12 . The computer-implementable method of claim 1 , wherein, in response to the pre-trained decision rule rendering the positive outcome, the method further comprises:

restricting, by the at least one processor, access to the respective domain name of the given one of the refined plurality of DNS queries, for every network device included into the network infrastructure;

generating, by the at least one processor, a notification of the respective domain name being malicious;

generating, by the at least one processor, a template email for notifying interested parties about the respective domain name being malicious; and

adding the respective domain name to a blacklist of domain names.

13 . A computing device for identifying, in a network infrastructure, compromised network devices that are using DNS tunneling, the computing device being located within the network infrastructure, the computing device comprising at least one processor communicatively coupled to the network infrastructure and a non-transitory computer-readable memory storing instructions, the at least one processor, upon executing the instructions, being configured to:

receive traffic generated by each network device of a plurality of network devices located within the network infrastructure;

identify, from the traffic, a plurality of DNS queries having been generated by respective network devices of the plurality of network devices located within the network infrastructure,

each one of the plurality DNS queries having been transmitted outside the network infrastructure;

determine, in the plurality of DNS queries, presence of DNS queries to predetermined types of domain names, generation of which, by a respective network device of the network infrastructure, is non-indicative of the respective network device being compromised,

the predetermined types of domain names including at least low entropy domain names;

remove, from the plurality of DNS queries, the DNS queries to the predetermined types of domain names, thereby generating a refined plurality of DNS queries;

generate, for a given one of the refined plurality of DNS queries, a respective set of suspiciousness features, the respective set of suspiciousness features comprising:

statistics of network interactions of the respective network device having transmitted the given one of the refined plurality of DNS queries,

the statistics of network interactions including: (i) for protocols above a Transmission Control Protocol (TCP), a number of TCP sessions; and (ii) for User Datagram Protocol (UDP), a number of datagrams over a predetermined period after the respective network device most recently sent the given one of the refined plurality of DNS queries;

apply, to the respective set of suspiciousness features associated with the given one of the refined plurality of DNS queries, a pre-trained decision rule,

the pre-trained decision rule having been trained to determine, based on the respective set of suspiciousness features, whether a respective domain name associated with the given one of the refined plurality of DNS queries is malicious;

in response to the pre-trained decision rule rendering a positive outcome, increase a respective penalty score for each network device of the network infrastructure having transmitted the given one of the refined plurality of DNS queries;

in response to the respective penalty score associated with a given network device exceeding a predetermined penalty score threshold, identify the given network device as being compromised; and

generate a warning notification for transmission thereof about the given network device being compromised.

Assignments (3)

Priority Claims (1)

Continuity (1)

References Cited (287)