In-app behavior-based attack detection
Architectures and techniques for in-app behavior detection. A behavior detection agent within an application running on a hardware computing device captures events within the application. The events are inputs received from one or more sources external to the application. The behavior detection agent generates an event stream from the captured events. The behavior detection agent analyzes the event stream for significant feature frequencies and associations corresponding to one or more attack profiles. The behavior detection agent initiates an attack response in response to finding one or more significant feature frequencies and associations. The attack response comprises at least changing an operational configuration of the application.
1. A non-transitory computer-readable medium having stored thereon instructions that, when executed by one or more processors, are configurable to cause the one or more processors to:
receive, with a behavior detection agent in a hardware computing platform, a set of feature vectors, each feature vector having at least timing and event information for a corresponding event;
apply, with the behavior detection agent, to the set of feature vectors a sub-optimal, multi-phase analysis to determine probabilities that selected features observed as positive or negative will occur;
select a subset of features, with the behavior detection agent, based on the determined probabilities;
generate, from the selected subset of features with the behavior detection agent, at least one sub-optimal function having an ordered set of feature detection rules;
formulate, with the behavior detection agent, an attack response based on the at least one sub-optimal function to be utilized in an attack pattern detection mechanism; and
apply the attack response within at least the hardware computing platform via the behavior detection agent.
2. The non-transitory computer-readable medium of claim 1 wherein the multi-phase analysis comprises pattern induction via sub-exhaustive satisfiability tests.
3. The non-transitory computer-readable medium of claim 1 wherein the set of feature vectors corresponds to captured events in an event stream.
4. The non-transitory computer-readable medium of claim 1 , wherein at least one phase of the analysis comprises use of prior results to determine one or more probabilities that selected features observed as positive or negative in training data can be used to select a subset of features for the feature vector.
5. The non-transitory computer-readable medium of claim 4 wherein significant feature frequencies and associations is described by a Boolean function based on known positive and negative instances of selected conditions.
6. The non-transitory computer-readable medium of claim 5 wherein the Boolean function is generated by:
analyzing prior results to determine, for each of the significant features, determining an intersection of known positives and detected positives, and an intersection of detected positives and known negatives; and
producing a sub-optimal Boolean function in which detection rules are ranked by satisfiability rates.
7. The non-transitory computer-readable medium of claim 4 wherein at least a portion of the prior results are acquired from one or more log files.
8. The non-transitory computer-readable medium of claim 5 wherein the significant feature frequencies and associations comprises at least a timing between a first event and a second event.
9. A system comprising:
a memory system having one or more memory devices;
one or more hardware processors coupled with the memory system, the one or more hardware processors configurable to receive a set of feature vectors, each feature vector having at least timing and event information for a corresponding event, to apply to the set of feature vectors a sub-optimal, multi-phase analysis to determine probabilities that selected features observed as positive or negative will occur, to select a subset of features based on the determined probabilities, to generate, from the selected subset of features, at least one sub-optimal function having an ordered set of feature detection rules, to formulate an attack response based on the at least one sub-optimal function to be utilized in an attack pattern detection mechanism, and to apply the attack response within at least the hardware computing platform.
10. The system of claim 9 wherein the multi-phase analysis comprises pattern induction via sub-exhaustive satisfiability tests.
11. The system of claim 9 wherein the set of feature vectors corresponds to captured events in an event stream.
12. The system of claim 9 , wherein at least one phase of the analysis comprises use of prior results to determine one or more probabilities that selected features observed as positive or negative in training data can be used to select a subset of features for the feature vector.
13. The system of claim 12 wherein significant feature frequencies and associations is described by a Boolean function based on known positive and negative instances of selected conditions.
14. The system of claim 13 wherein the Boolean function is generated by:
analyzing prior results to determine, for each of the significant features, determining an intersection of known positives and detected positives, and an intersection of detected positives and known negatives; and
producing a sub-optimal Boolean function in which detection rules are ranked by satisfiability rates.
15. The system of claim 12 wherein at least a portion of the prior results are acquired from one or more log files.
16. The system of claim 13 wherein the significant feature frequencies and associations comprises at least a timing between a first event and a second event.
17. A non-transitory computer-readable medium having stored thereon instructions that, when executed by one or more processors, are configurable to cause the one or more processors to:
receive, with a behavior detection agent in a hardware computing platform, a set of feature vectors, each feature vector having at least timing and event information for a corresponding event;
apply, with the behavior detection agent, to the set of feature vectors a sub-optimal, multi-phase analysis to determine probabilities that selected features observed as positive or negative will occur;
select a subset of features, with the behavior detection agent, based on the determined probabilities;
generate, from the selected subset of features with the behavior detection agent, at least one sub-optimal function having an ordered set of feature detection rules;
formulate, with the behavior detection agent, an attack response based on the at least one sub-optimal function to be utilized in an attack pattern detection mechanism; and
apply the attack response within at least the hardware computing platform via the behavior detection agent.
18. The method of claim 17 wherein the multi-phase analysis comprises pattern induction via sub-exhaustive satisfiability tests.
19. The method of claim 17 wherein the set of feature vectors corresponds to captured events in an event stream.
20. The method of claim 17 , wherein at least one phase of the analysis comprises use of prior results to determine one or more probabilities that selected features observed as positive or negative in training data can be used to select a subset of features for the feature vector.