As discussed in our last post, evolved threat detection’s first step is gathering internal and external security data. Once you have the data aggregated you need to analyze it to look for indications that you have compromised devices and/or malicious activity within your organization.
Know Your Assets
You know the old business adage: you can’t manage it if you can’t see it. In security monitoring parlance, you need to discover new assets – and changes to existing ones – to monitor them, and ultimately to figure out when a device has been compromised. A key aspect to threat detection remains discovery. The enemy of the security professional is surprise, so it is essential to always be aware of network topology and devices on the network. All devices, especially those pesky rogue wireless access points and other mobile devices, provide attack surface to adversaries.
How can you make sure you are continuously discovering these devices? You scan your address space. Of course there is active scanning, but that runs periodically. To fill in between active scans, passive scanning watches network traffic streaming by to identify devices you haven’t seen or which have changed. Once a device is identified passively, you can launch an active scan to figure out what it’s doing (and whether it is legitimate). Don’t forget to discover your entire address space – which means both IPv4 and IPv6.
Most discovery efforts focus on PCs and servers on the internal network. But that may not be enough anymore; it is typically endpoints that end up compromised, so you might want to discover both full computers and mobile devices. Finally, you will need to figure out how to discover assets in your cloud computing environments. This requires integration with cloud consoles to ensure you know about new cloud-based resources and can monitor them appropriately.
After you have a handle on the devices within your environment, the next step is to classify them. We recommend a simple classification, involving roughly 4 groupings. The most important bucket includes critical devices with access to private information and/or valuable intellectual property. Next look for devices behaving maliciously. These devices may not have sensitive information, but adversaries can move laterally from compromised devices to critical devices. Then you have dormant devices, which may have connected to a command and control infrastructure but aren’t currently doing anything malicious. Finally, there are all the other devices which aren’t doing anything suspicious – which you likely don’t have time to worry about. We introduced this categorization in the Network-based Threat Detection series – check it out if you want more detail.
Finally, we continue to harp on the criticality of a consistent process for threat detection. This includes discovery and classification. As with data collection, your technology environment is dynamic, so what you saw 10 minutes ago will have changed by 20 minutes in the future – or sooner. You need a strong process to ensure you always understand what is happening in your environment.
The C Word
Correlation has always been a challenge for security folks. It’s not because the math doesn’t work. Math works just fine. Event correlation has been a challenge because you needed to know what to look for at a very granular level. Given the kinds of attacks and advanced adversaries many organizations face, you cannot afford to count on knowing what’s coming, so it’s hard to find new and innovative attacks via traditional correlation. This has led to generally poor perceptions of SIEMs and IDS/IPS.
But that doesn’t meant correlation is useless for security. Quite the opposite. Looking for common attributes, and linking events together into meaningful models of possible attacks, provides a meaningful way to investigate security events. And you don’t want to succumb to the same attacks over and over again, so it is still important to look for indicators of attacks that have been used against you. Even better if you can detect indicators reported by other organizations, via threat intelligence, and avoid those attacks entirely.
Additionally you can (and should) stage out a number of reasonable attack patterns via threat modeling to look for common attacks. In fact, your vendor or service provider’s research team has likely built in some of these common patterns to kickstart your efforts at building out correlation rules, based on their research. These research teams also keep their correlation rules current, based on what they see in the wild.
Of course you can never know all possible attacks. So you also need to apply behavioral and other advanced analytical techniques to catch attacks you have not seen.
Looking for Outliers
Technology systems have typical activity patterns. Whether network traffic, log events, transactions, or any other kind of data source, you can establish an activity profile for how systems normally behave. Once the profile is established you look for anomalous activity, or outliers, that may represent malicious activity. Theses outliers could be anything, from any data source you collect.
With a massive trove of data, you can take advantage of advanced “Big Data” analytics (no, we don’t like that overly vague term). New technologies can reduce a huge amount of data to scan for abnormal activity patterns. You need an iterative process to refine thresholds and baseline over time. Yes, that means ongoing care and feeding of your security analytics. Activity evolves over time, so today’s normal might be anomalous in a month.
Setting up these profiles and maintaining the analytics typically requires advanced skills. The new term for these professionals is data scientists. Yes, it’s a shiny term, and practitioners are expensive. But a key aspect of detecting threats is looking for outliers, and that requires data scientists, so you’ll need to pay up. Just ensure you have sufficient resources to investigate alerts coming from your analytics engine, because if you aren’t staffed to triage and validate alerts, you waste the benefit of earlier threat detection.
Alternatively, organizations without these sophisticated internal resources should consider allowing a vendor or service provider to update and tune their correlation rules and analytics for detection. This is especially helpful as organizations embrace more advanced analytics without internal data scientists to run the math.
Visualization and Drill down
Given the challenges of finding skilled resources for triage and validation, you’ll need to supplement internal skills with technology-accelerated functions. That means better visualization, and a built-in workflow to validate and triage alerts. You want a straightforward graphical metaphor to help categorize and prioritize alerts, and then a way to dig into an alert to really understand what is happening and identify root cause.
The only way to get a feel for whether a visual metaphor will work for you is to actually use it. That’s why a proof of concept (PoC) is so important when looking at detection technologies and services. You’ll be able to pump some of your data into the tool, generate alerts, and validate them as you would in a production deployment. Even better, you’ll have skilled resources from the vendor or channel partner to help stand up the system, perform initial configuration, and work through some alerts. Take advantage of these resources to kickstart your efforts.
Standalone analytics can work, especially for very specialized use cases such as large financial institutions addressing the insider threat, we believe a more generic detection platform can make a significant impact in resource-constrained environments. Not having to perform manual triage and validation of issues can save a ton of time and supplement your internal skill sets, especially if you leverage a vendor’s security research and/or threat intelligence services.
So another key criteria for evolving threat detection is flexible integration with additional security data sources, emerging analytic techniques, advanced visualization engines, and operational workflow tools. Over time we expect the threat detection capability to morph into the core security monitoring platform collecting internal security data, absorbing threat intelligence from a number of external sources, providing analytics to detect attacks, and ultimately sending information on to operational systems and controls to change the environment.
Next we will wrap up this series with a Quick Wins scenario, presenting this theory in the context of an attack to see how evolved threat detection works in practice.