Here are some answers to frequently asked questions about detections.
- How are detections different from alerts?
- What is a risk score?
- Why can't I view source device details for a detection?
- How far back are detections found?
- Can I connect to the Machine Learning Service through a proxy?
- What data is sent from the ExtraHop system to the cloud-based Machine Learning Service?
- How secure are detections?
- How do I add a new or updated license for the Machine Learning Service to my ExtraHop system?
- After my Machine Learning Service license expires, can I still view my previous detections?
Alerts and detections are similar in that they both provide information about conditions on your network. The following table describes how they differ.
|How are they generated?||By conditions you define through Alert settings. You can configure trend, threshold, or detection alerts.||Automatically observed from your network data by the ExtraHop Machine Learning Service.|
|How do I view them?||Click Alerts from the top menu of the Web UI.||Click Detections from the top menu of the Web UI.|
|How do I set up email notifications?||After an email server is configured in the Admin UI, you can set up email notification settings for any alert.||After an email server is configured in the Admin UI, you can configure a detection alert and then set up email notifications.|
|What are the benefits?||You decide which business critical devices and services to monitor and determine the level of change that generates notification.||Notable changes to your network behavior are automatically surfaced. By providing feedback for detections, you help the Machine Learning Service algorithm better understand your network.|
|What are the challenges?||As your network changes, your configuration might become outdated or require continual maintenance of alert configurations.||Several security detections can be identified immediately after you connect to the ExtraHop Machine Learning Service; however, traffic must be stored on the Discover appliance for two weeks before the algorithm understands typical patterns of behavior for your network traffic and can identify performance detections.|
A risk score indicates the severity of a detection and is calculated based on the likelihood of an attack, the difficulty of exploiting the detection, and the level of impact to your operations.
Risk scores are grouped into one of the following color-coded severity levels:
- Red = 80-99
- Orange = 31-79
- Yellow = 1-30
Risk scores were added to detections in Reveal(x) Summer 2018. If a detection was identified in a previous version, the risk score is unavailable for that detection.
No risk score is displayed for an individual detection if a score has not been evaluated and defined for that detection.
If the source of a detection is a device that hasn't been discovered by the ExtraHop system, the detection only shows the IP address and hostname of the device, if available. You can hover over the undiscovered device to see the geolocation of the IP address and a link to the ARIN Whois website.
Machine-learning detections are identified one week back from the time the service is connected. The service then identifies all new detections moving forward.
Note that the Machine Learning Service requires four weeks (28 days) of data to calculate an expected range of metric values. The expected range represents normal network behavior. Data processing is typically completed within a few hours.
In ExtraHop 7.0 and later, the Machine Learning Service supports implicit and explicit proxies. The proxy requires that DNS resolve all *.extrahop.com domains, and the outbound 443 port is open to all IP addresses on the internet. These settings are implemented on the firewall for the proxy's source IP address.
For more information on configuring an explicit proxy, see Connect to ExtraHop Cloud Services through a proxy.
The Machine Learning Service takes advantage of the unique processing capabilities of the ExtraHop system to "pre-process" wire data for hundreds of metrics on-premise. The ExtraHop system encrypts metric values and IP addresses that are sent to the Machine Learning Service. The ExtraHop system does not send custom metrics or sensitive data such as file names, strings, or payloads.
Detections are designed to be secure from end-to-end. Unlike a typical SaaS solution, detections do not ingest payloads, file names, strings, or other data categories that might contain sensitive information. Sensitive data remains on-premise and under your control. The ExtraHop Machine Learning Service has received the SOC 2, Type 1 compliance certification.
If you purchased a new ExtraHop system that includes a license for the Machine Learning Service, you will receive an email with a new product key that must added to your appliance. Follow the instructions to register your appliance.
If you have added a license for the Machine Learning Service, your updated license is automatically added to your ExtraHop system, but must still be applied. Follow the instructions to apply an updated license.