Detection of information concealment and mitigation via introduction of deliberate variance
Network information is obtained for a plurality of discrete information units transmitted by a transmitting entity. The network information is descriptive of features of network events that occurred for the discrete information units. The network information is processed with a machine-learned hidden information detection model to obtain a prediction output indicating that the features of the network events that occurred for the discrete information units are modulated to conceal hidden information. Based on the prediction output, a variance addition process is caused to be performed for a second plurality of discrete information units to be transmitted by the transmitting entity. The variance addition process causes variance to be added to features of future network events for at least some of the second plurality of discrete information units.
1 . A computer-implemented method, comprising:
obtaining, by a computing system comprising one or more computing devices, network information for a plurality of discrete information units transmitted by a transmitting entity, wherein the network information is descriptive of temporal features of network events that occurred for the plurality of discrete information units;
processing, by the computing system, the network information with a machine-learned hidden information detection model to obtain a prediction output indicating that the temporal features of the network events that occurred for the plurality of discrete information units are modulated to conceal hidden information; and
based on the prediction output, causing, by the computing system, a variance addition process to be performed for a second plurality of discrete information units to be transmitted by the transmitting entity, wherein the variance addition process causes deliberate variance to be added to temporal features of future network events for at least some of the second plurality of discrete information units.
2 . The computer-implemented method of claim 1 , wherein the network events that occurred for the plurality of discrete information units comprise at least one of:
receipt of a discrete information unit of the plurality of discrete information units by a receiving entity;
occurrence of a hop from one network segment to another network segment; or
transmission of a discrete information unit of the plurality of discrete information units by the transmitting entity.
3 . The computer-implemented method of claim 1 , wherein, for each of the plurality of discrete information units, the temporal features of the network events that are modulated to conceal the hidden information comprise at least one of:
a timestamp indicative of a time at which a network event occurred;
a hop limit that controls a maximum number of hop network events allowed for a discrete information unit;
a sequence number;
a packet identifier; or
an address that identifies a network entity.
4 . The computer-implemented method of claim 1 , wherein processing the network information with the machine-learned hidden information detection model further comprises:
obtaining, by the computing system, information indicative of the second plurality of discrete information units for transmission by the transmitting entity.
5 . The computer-implemented method of claim 1 , wherein the plurality of discrete information units respectively comprises a plurality of packets.
6 . The computer-implemented method of claim 5 , wherein obtaining the network information comprises:
obtaining, by the computing system, the network information for the plurality of packets transmitted by the transmitting entity, wherein the network information is descriptive of:
features of the plurality of packets; and
features of the network events that occurred for the plurality of packets.
7 . The computer-implemented method of claim 6 , wherein processing the network information with the machine-learned hidden information detection model to obtain the prediction output comprises:
extracting, by the computing system, one or more feature tuples for one or more packets of the plurality of packets, wherein each of the one or more feature tuples is indicative of at least one of:
features of the plurality of packets; or
temporal features of the network events that occurred for the plurality of packets; and
processing, by the computing system, the one or more feature tuples with the machine-learned hidden information detection model to obtain the prediction output indicating that features indicated by the one or more feature tuples are modulated to conceal the hidden information.
8 . The computer-implemented method of claim 7 , wherein processing the one or more feature tuples with the machine-learned hidden information detection model comprises:
processing, by the computing system, the one or more feature tuples with the machine-learned hidden information detection model to obtain the prediction output, wherein the prediction output identifies one or more modulated features from features included in the one or more feature tuples, and wherein the one or more modulated features are modulated to conceal the hidden information.
9 . The computer-implemented method of claim 8 , wherein a first modulated feature of the one or more modulated features is associated with a first packet of the one or more packets, and wherein the first modulated feature comprises:
a network event timestamp for a network event that occurred for the first packet;
an originating process name for a first process from which the first packet originates;
an originating process identifier for the first process;
an identifier for a calling entity that calls the first process prior to creation of the first packet;
a source Internet Protocol (IP) address;
a source port identifier;
a destination IP address;
a destination port identifier;
a service type associated with the first process;
a transmission protocol associated with the first packet; or
geolocation data.
10 . The computer-implemented method of claim 8 , wherein the method further comprises:
evaluating, by the computing system, a loss function that evaluates a difference between the prediction output and a ground truth prediction output; and
modifying, by the computing system, values of parameters of the machine-learned hidden information detection model based on the loss function.
11 . The computer-implemented method of claim 8 , wherein processing the one or more feature tuples with the machine-learned hidden information detection model comprises:
processing, by the computing system, the one or more feature tuples with the machine-learned hidden information detection model to obtain the prediction output, wherein the prediction output identifies the one or more modulated features from the features included in the one or more feature tuples, and wherein the one or more modulated features comprise sequence identifiers for the one or more packets.
12 . The computer-implemented method of claim 11 , wherein causing the variance addition process to be performed for the second plurality of discrete information units comprises:
causing, by the computing system, the variance addition process to be performed for the second plurality of discrete information units, wherein the variance addition process causes variance to be respectively added to one or more sequence identifiers for one or more discrete information units of the second plurality of discrete information units.
13 . The computer-implemented method of claim 8 , wherein processing the one or more feature tuples with the machine-learned hidden information detection model comprises:
processing, by the computing system, the one or more feature tuples with the machine-learned hidden information detection model to obtain the prediction output, wherein the prediction output identifies the one or more modulated features from the features included in the one or more feature tuples, and wherein the one or more modulated features comprises transmission times for the one or more packets.
14 . The computer-implemented method of claim 13 , wherein causing the variance addition process to be performed for the second plurality of discrete information units comprises:
causing, by the computing system, the variance addition process to be performed for the second plurality of discrete information units, wherein the variance addition process causes a variable delay to be added during transmission of the second plurality of discrete information units.
15 . The computer-implemented method of claim 8 , wherein processing the one or more feature tuples with the machine-learned hidden information detection model comprises:
processing, by the computing system, the one or more feature tuples with the machine-learned hidden information detection model to obtain the prediction output, wherein the prediction output identifies the one or more modulated features from the features included in the one or more feature tuples, and wherein the one or more modulated features comprise hop limits for the one or more packets.
16 . The computer-implemented method of claim 15 , wherein causing the variance addition process to be performed for the second plurality of discrete information units comprises:
causing, by the computing system, the variance addition process to be performed for the second plurality of discrete information units, wherein the variance addition process modifies a hop limit of one or more discrete information units of the second plurality of discrete information units.
17 . A computing system, comprising:
one or more processors; and
one or more non-transitory computer-readable media that store instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations comprising:
obtaining network information for a plurality of packets transmitted by a transmitting entity, wherein the network information is descriptive of temporal features of network events that occurred during transmission of the plurality of packets;
extracting one or more feature tuples for one or more packets of the plurality of packets, wherein each of the one or more feature tuples is indicative of the temporal features of the network events that occurred for the plurality of packets;
processing the one or more feature tuples with a machine-learned hidden information detection model to obtain a prediction output that identifies one or more modulated features from features included in the one or more feature tuples, and wherein the one or more modulated features are modulated to conceal hidden information;
obtaining information indicative of a second plurality of packets for transmission by the transmitting entity; and
based on the prediction output, causing a variance addition process to be performed for the second plurality of packets, wherein the variance addition process causes deliberate variance to be added to the temporal features of future network events for at least some of the second plurality of packets.
18 . The computing system of claim 17 , wherein the network events that occurred for the plurality of packets comprises at least one of:
receipt of a discrete information unit of the plurality of packets by a receiving entity;
occurrence of a hop from one network segment to another network segment; or
transmission of a packet of the plurality of packets by the transmitting entity.
19 . The computing system of claim 17 , wherein the one or more modulated features comprise at least one of:
a timestamp indicative of a time at which a network event occurred;
a hop limit that controls a maximum number of hop network events allowed for a discrete information unit;
a sequence number;
a packet identifier; or
an address that identifies a network entity.
20 . One or more non-transitory computer-readable media that store instructions that, when executed by one or more processors, cause the one or more processors to perform operations, the operations comprising:
obtaining network information for a plurality of discrete information units transmitted by a transmitting entity, wherein the network information is descriptive of temporal features of network events that occurred for the plurality of discrete information units;
processing the network information with a machine-learned hidden information detection model to obtain a prediction output indicating that the temporal features of the network events that occurred for the plurality of discrete information units are modulated to conceal hidden information; and
based on the prediction output, causing a variance addition process to be performed for the transmitting entity, wherein the variance addition process causes deliberate variance to be added to temporal features of future network events for future transmissions of discrete information units by the transmitting entity.