Anomaly and ransomware detection
Some examples relate generally to computer architecture software for information security and, in some more particular aspects, to machine learning based on changes in snapshot metadata for anomaly and ransomware detection in a file system.
1 . A system, comprising:
a storage device configured to store one or more snapshots of a primary machine; and
one or more processors in communication with the storage device and a production system, the one or more processors configured to perform one or more anomaly and ransomware detection operations comprising:
taking a first snapshot and a second snapshot of the primary machine;
identifying one or more changes in snapshot-based metadata based on the first snapshot and the second snapshot;
training, in accordance with a two-stage machine learning analysis, a first machine-learning model that compares the one or more changes in the snapshot-based metadata occurring over a time duration to a threshold quantity of changes indicative of an anomalous event and a second machine-learning model different from the first machine-learning model that performs an entropy scan of the one or more changes in the snapshot-based metadata to identify whether respective entropies associated with the one or more changes in the snapshot-based metadata comprise a ransomware encryption event, wherein the training of the first machine-learning model and the second machine-learning model is further based on training data derived from the one or more changes in the snapshot-based metadata;
detecting an anomaly from a new snapshot of the primary machine using the first machine-learning model at a first stage of the two-stage machine learning analysis; and
passing, in response to detecting the anomaly, the new snapshot of the primary machine through the second machine-learning model to detect whether the detected anomaly is a ransomware encryption anomaly at a second stage of the two-stage machine learning analysis.
2 . The system of claim 1 , wherein the storage device is a backup storage device and the one or more changes in the snapshot-based metadata are sourced from a backup system that includes the backup storage device.
3 . The system of claim 1 , wherein the one or more anomaly and ransomware detection operations are performed without impacting the production system.
4 . The system of claim 1 , wherein training the first machine-learning model further comprises:
analyzing the one or more changes in the snapshot-based metadata over the time duration, wherein the one or more changes comprise a threshold number of file creations, a threshold number of file deletions, a threshold number of file modifications, or any combination thereof.
5 . The system of claim 1 , wherein the first machine-learning model and the second machine-learning model comprise an anomaly model, an encryption model, or both.
6 . The system of claim 1 , wherein training the second machine-learning model comprises:
performing the entropy scan of the one or more changes in the snapshot-based metadata to identify
whether an event comprises the ransomware encryption event based at least in part on the respective entropies.
7 . A method at an anomaly and ransomware detection system, the anomaly and ransomware detection system including one or more processors in communication with a storage device and a production system, the one or more processors configured to perform one or more anomaly and ransomware detection operations comprising:
taking a first snapshot and a second snapshot of a primary machine;
identifying one or more changes in snapshot-based metadata based on the first snapshot and the second snapshot;
training, in accordance with a two-stage machine learning analysis, a first machine-learning model that compares the one or more changes in the snapshot-based metadata occurring over a time duration to a threshold quantity of changes indicative of an anomalous event and a second machine-learning model different from the first machine-learning model that performs an entropy scan of the one or more changes in the snapshot-based metadata to identify whether respective entropies associated with the one or more changes in the snapshot-based metadata comprise a ransomware encryption event, wherein the training of the first machine-learning model and the second machine-learning model is further based on training data derived from the one or more changes in the snapshot-based metadata;
detecting an anomaly from a new snapshot of the primary machine using the first machine-learning model at a first stage of the two-stage machine learning analysis; and
passing, in response to detecting the anomaly, the new snapshot of the primary machine through the second machine-learning model to detect whether the detected anomaly is a ransomware encryption anomaly at a second stage of the two-stage machine learning analysis.
8 . The method of claim 7 , wherein the storage device is a backup storage device and the one or more changes in the snapshot-based metadata are sourced from a backup system that includes the backup storage device.
9 . The method of claim 7 , wherein the one or more anomaly and ransomware detection operations are performed without impacting the production system.
10 . The method of claim 7 , wherein training the first machine-learning model further comprises:
analyzing the one or more changes in the snapshot-based metadata over the time duration, wherein the one or more changes comprise a threshold number of file creations, a threshold number of file deletions, a threshold number of file modifications, or any combination thereof.
11 . The method of claim 7 , wherein the first machine-learning model and the second machine-learning model comprise an anomaly model, an encryption model, or both.
12 . The method of claim 7 , wherein training the second machine-learning model comprises:
performing the entropy scan of the one or more changes in the snapshot-based metadata to identify
whether an event comprises the ransomware encryption event based at least in part on the respective entropies.
13 . A non-transitory, machine-readable medium storing instructions which, when read by a machine, cause the machine to perform one or more anomaly and ransomware detection operations comprising:
taking a first snapshot and a second snapshot of a primary machine;
identifying one or more changes in snapshot-based metadata based on the first snapshot and the second snapshot;
training, in accordance with a two-stage machine learning analysis, a first machine-learning model that compares the one or more changes in the snapshot-based metadata occurring over a time duration to a threshold quantity of changes indicative of an anomalous event and a second machine-learning model different from the first machine-learning model that performs an entropy scan of the one or more changes in the snapshot-based metadata to identify whether respective entropies associated with the one or more changes in the snapshot-based metadata comprise a ransomware encryption event, wherein the training of the first machine-learning model and the second machine-learning model is further based on training data derived from the one or more changes in the snapshot-based metadata;
detecting an anomaly from a new snapshot of the primary machine using the first machine-learning model at a first stage of the two-stage machine learning analysis; and
passing, in response to detecting the anomaly, the new snapshot of the primary machine through the second machine-learning model to detect whether the detected anomaly is a ransomware encryption anomaly at a second stage of the two-stage machine learning analysis.
14 . The non-transitory, machine-readable medium of claim 13 , wherein the one or more changes in the snapshot-based metadata are sourced from a backup system that includes a backup storage device.
15 . The non-transitory, machine-readable medium of claim 13 , wherein the one or more anomaly and ransomware detection operations are performed without impacting a production system.
16 . The non-transitory, machine-readable medium of claim 13 , wherein training the first machine-learning model further comprises:
analyzing the one or more changes in the snapshot-based metadata over the time duration, wherein the one or more changes comprise a threshold number of file creations, a threshold number of file deletions, a threshold number of file modifications, or any combination thereof.
17 . The non-transitory, machine-readable medium of claim 13 , wherein the first machine-learning model and the second machine-learning model comprise an anomaly model, an encryption model, or both.
18 . The non-transitory, machine-readable medium of claim 13 , wherein training the second machine-learning model comprises:
performing the entropy scan of the one or more changes in the snapshot-based metadata to identify
whether an event comprises the ransomware encryption event based at least in part on the respective entropies.