Analysis of endpoint detect and response data
A computing apparatus to provide endpoint detect and response (EDR) filtering to an enterprise, including: a processor and memory; a network interface; a network protocol to communicatively couple to a data source via the network interface; and instructions encoded within the memory to provide an EDR filtering pipeline to receive an unfiltered EDR stream via the network interface, extract an EDR record from the EDR stream, and apply a hash to the EDR record to determine that the EDR record is uncommon in context of the enterprise; and a decorator module to decorate the EDR record for in-depth analysis.
1. A computing apparatus to provide endpoint detect and response (EDR) filtering to an enterprise, comprising:
a processor and memory;
a network interface; and
instructions encoded within the memory to:
provide an EDR filtering pipeline to receive an unfiltered EDR stream via the network interface, wherein the unfiltered EDR stream comprises n elements; and
build a filtered EDR stream of m elements, comprising extracting EDR records from the unfiltered EDR stream, and applying a hash to the EDR records to identify uncommon EDR records that are uncommon in context of the enterprise, and selecting the uncommon EDR records for in-depth analysis, wherein m is at least one order of magnitude smaller than n; and
a decorator module to decorate the uncommon EDR records for in-depth analysis, including human analysis.
2. The computing apparatus of claim 1 , further comprising an interface to provide the uncommon EDR records to a security operations center (SOC) for in-depth analysis.
3. The computing apparatus of claim 2 , wherein the interface comprises a user interface for a human security expert.
4. The computing apparatus of claim 3 , wherein the user interface comprises controls for the human security expert to adjust a sensitivity of determining that the EDR record is uncommon in context of the enterprise.
5. The computing apparatus of claim 2 , wherein the interface comprises an interface to a cooperative human/machine analysis operation within the SOC.
6. The computing apparatus of claim 1 , wherein the hash comprises a MinHash algorithm.
7. The computing apparatus of claim 6 , wherein applying the MinHash algorithm comprises identifying command line operations within the EDR records, and tokenizing the command line operations on a token.
8. The computing apparatus of claim 7 , wherein the token is a space.
9. The computing apparatus of claim 8 , wherein applying the MinHash algorithm further comprises assigning a numerical value to each token, and computing an overall signature for the MinHash algorithm from the numerical values.
10. The computing apparatus of claim 1 , wherein applying the hash to the EDR records further comprises performing locality-sensitive hashing (LSH) on the EDR records to bucketize a signatures of the EDR records.
11. The computing apparatus of claim 1 , wherein the instructions further comprise instructions to apply a machine learning model to the unfiltered EDR stream to build the filtered EDR stream.
12. The computing apparatus of claim ii, wherein the instructions comprise an in-memory machine learning model.
13. The computing apparatus of claim ii, wherein the instructions further comprise instructions to synthesize a version of the machine learning model, and export the synthesized version to one or more endpoint devices via the network interface.
14. The computing apparatus of claim 13 , further comprising a software network protocol to connect to a publish/subscribe bus.
15. One or more tangible, non-transitory computer-readable storage media having stored thereon executable instructions to:
communicatively couple to a network service;
receive a stream of endpoint detect and response (EDR) records via the network service, the stream comprising n EDR records;
select uncommon EDR records for analysis, wherein selecting comprises identifyinga command line operation of an EDR record, computing a MinHash of the EDR record$ and applying a machine learning commonality model to the MinHash to determine that the EDR record is behaviorally uncommon for an enterprise, wherein selecting yields a set of uncommon EDR records comprising m EDR records, wherein m is at least one order of magnitude less than n; and
mark the uncommon EDR records as uncommon and select the uncommon EDR records for in-depth analysis, including human analysis.
16. The one or more tangible, non-transitory computer-readable media of claim 15 , wherein the instructions are further to receive from a security operations center (SOC) an indication that the EDR record exhibits malicious or negligent behavior, and associate a malware reputation with the EDR record.
17. The one or more tangible, non-transitory computer-readable media of claim 15 , wherein computing the MinHash comprises identifying a command line operation within the EDR record, and tokenizing the command line operation.
18. A method of filtering endpoint detect and response (EDR) records within an enterprise, comprising:
communicatively coupling to an enterprise network;
receiving a stream of n endpoint detect and response (EDR) records from a plurality of enterprise endpoint devices via the enterprise network;
selecting a set of m uncommon EDR records from the stream of EDR records, wherein m is at least one order of magnitude less than n, wherein selecting the set of uncommon EDR records comprises tokenizing a command line string within an EDR record of the stream of EDR records, assigning a numerical value to each token of the tokenized command line string, computing a MinHash signature for the EDR record from the numerical values, applying locality-sensitive hashing to the MinHash signature to bucketize the MinHash signature, and determining that the MinHash, signature marks the EDR record as exhibiting uncommon behavior for the enterprise; and
marking the uncommon EDR records for in-depth analysis, including human analysis.
19. The method of claim 18 , further comprising receiving from a security operations center (SOC) an indication that the EDR record exhibits malicious or negligent behavior, and associating a malware reputation with the EDR record.
20. The method of claim 19 , further comprising detecting an additional EDR record with a signature that identifies the additional EDR record as being similar to the EDR record of the stream of EDR records, and assigning to the additional EDR record a reputation of being suspicious or malicious.