This article is contributed. See the original author and article here.
With so many external cyber threats facing Government agencies, it can be easy to overlook risks from insiders that may have malicious objectives or that may make unintentional but serious mistakes. Digital transformation and modernization of Government agencies have enabled new efficiencies and created an exponential increase in data that is stored and processed digitally. As an agency’s data becomes increasingly digital, many of the physical security and privacy risks associated with that data become digital as well. Cyber threats from entities outside of an organization are often well-known as a result of press attention and industry threat intelligence sharing; however, cyber threats from individuals inside an organization are less well-known despite being fairly common. A study by Delloitte found that 59% of employees who leave an organization take sensitive data with them and 25% of employees have used email to exfiltrate sensitive data from an organization.
What is an insider?
Let’s start with the definition of an insider – an insider is a person working within a group or organization, often privy to information unavailable to others. The key aspect of this definition with security implications is that there is an expectation that this person will have additional direct or indirect access to information that others would not. As a result, a simple “deny access” approach that might work as an outsider strategy cannot be applied to insiders. Third parties such as contractors, subcontractors, vendors, and suppliers are also considered insiders when they receive additional access to people, devices, or information. Once an individual is authorized to access the information or system, defensive tactics move beyond the realm of simple access management into managing risk from authorized users – a realm defined by monitoring usage and behavior. Continuous monitoring is essential, but telemetry alone will not result in successful mitigation of risks from insiders. The difference between legitimate and illegitimate activity often boils down to discerning intent and context more so than assessing the legitimacy of a single isolated event.
What are the risks?
Insider risks span a broad range of possibilities including theft, data spillage, security control violations, compliance violations, espionage, sabotage, and workplace harassment or violence. Sufficiently addressing these risks requires having the right people and processes in place empowered by the right technology to glean insights and assess the possible risks at scale. Analysts and investigators need to be able to prioritize investigations based upon an incident’s full context, impact, and intent. To make meaningful assessments quickly they need to have all relevant information available at their fingertips and receive only those cases that have a sufficiently high degree of confidence.
Zero Trust is a strategy that has been gaining significant momentum in the public sector. One of the key assertions of a Zero Trust strategy is that a device or user should not be considered trusted just because it is operating on a trusted internal network. The same Zero Trust approach can also be applied to insider risks, by asserting that actions taken by insiders should not be considered trusted or “safe” just because of an individual’s position within an organization. Moving beyond reliance on simple allow/deny access controls towards continuous monitoring of activities and contextual risk is a proven Zero Trust methodology that can also advance insider risk reduction efforts. The additional telemetry and context in a Zero Trust implementation provide more opportunities to detect risky activity.
Using Predictive Analytics
Predictive analytics are essential to realizing a successful continuous monitoring program that can effectively mitigate risks from insider threats at scale. Taking a reactive or “forensic-only” approach by waiting until after the impact has been discovered opens an organization up to a massive and unnecessary level of risk. In contrast, a proactive approach can provide early warnings and mitigation opportunity prior to impact. A proactive approach becomes even more effective when it is made predictive through intentional planning and optimization for the most likely insider risk scenarios.
Step 1 – Gathering Intelligence
The first step towards predictive analytics is to research and obtain high quality threat intelligence which allows for refining the objectives of the continuous monitoring program and yields a high return on investment. Insider threat intelligence is not made up of network indicators or file hashes, but rather details of common tactics and techniques used by insiders. Insights from studying previous insider cases within your own agency or from a similar agency are especially useful for harvesting priority scenarios since past behavior can be a powerful indicator of future risk. In the absence of case studies, court records can also provide valuable insights into tactics and techniques used by insiders. Third–party threat intelligence from vendors is also a significant asset that can help surface global trends or tactics at an application/workload level. Carnegie Mellon University also maintains a publicly available set of insider threat resources. The outcome of this research activity should be the prioritization of specific scenarios that can reveal optimal detections of specific tactics and techniques based upon which ones are most likely to be used within your agency and those that would have the greatest impact if left undetected. For planning purposes, these scenarios should also be separated into the two major categories of intentional threats (malicious insider) and unintentional threats (insider who makes a mistake or is compromised by social engineering), to ensure adequate coverage for each.
Step 2 – Collecting Telemetry
The second step is to ensure that the necessary telemetry is available to enable direct or indirect monitoring of the priority scenarios. Identifying the core priority scenarios during the first step should drive the planning done during this step to reduce cost and ensure that the data being collected is needed and once collected becomes fully utilized. At Microsoft we use telemetry from various logs, activities, entity relationships, and alerts to build the Microsoft Intelligent Security Graph that provides threat insights and informs detections in our products.
Step 3 – Targeting Detections
The third step is to implement detections for each activity in the priority insider risk scenarios using the telemetry gathered in step two. To minimize complexity and maintenance of these detections, make use of workload-specific or activity-specific detections from vendors wherever possible such as Microsoft 365 Security solutions to monitor threats and actions in productivity workloads. Achieving a breadth of detections for each activity in the chain of each scenario over time is as important as achieving high quality individual activity detections. Detections that are small in scope, testable, and explainable generally are the best.
Step 4 – Machine Learning
The fourth step is to leverage pre-built machine learning (ML) capabilities such as those in Insider Risk Management whenever possible that are already tuned for your prioritized scenarios. You can also Build-Your-Own ML with Azure Sentinel. Azure is uniquely suited for building your own cyber ML with Sentinel’s entity extraction, cloud scale query capabilities, long-term data retention, notebooks for research, and Azure’s just-in-time compute availability for model training. We have found that using machine learning models is a necessity to address insider risk challenges at scale. Successful detection of insider threats often requires much more data over a greater time period than the detection of external threats. Without machine learning, detections become too complex for maintainable rules, queries over large time periods become memory and compute intensive, and maintaining performance becomes costly. Effective ML solutions for insider risk extend beyond context correlation or anomaly detection and are trained on specific threats from priority scenarios across sequences of activities and behaviors to predict an individual’s intent. Microsoft utilizes a comprehensive ML lifecycle to constantly research, engineer, and tune many different types and combinations of models for insider risk. For large organizations it is impossible to rely on anomaly detections or event correlation alone to surface risks. Without a predictive analytics approach the number of possible cases per month can easily range from thousands to millions.
Culture and Balance
Strategies to address risks from insiders also need to be balanced appropriately such that they do not disrupt an agency’s mission, reduce worker productivity, or undermine trust in the organization. Telemetry should be pervasive but not invasive, appropriately considering both employee privacy and organizational risk. A successful program also requires collaboration across HR, legal, and privacy teams to determine how to best address the organization’s priority risk scenarios. Having a broad base of stakeholders can ensure that a positive culture is maintained in the organization and that investigations are handled in a manner that has been established with broad agreement and support.
To find out more about how Microsoft is addressing insider threats with predicative analytics read a post on Risk Management on our Microsoft AI blog or watch Microsoft CISO Bret Arsenault discuss our Insider Risk Management strategy in detail on the Microsoft Mechanics channel. You can also find out more about how to integrate additional ML detections into your investigation workflow using the capabilities of Azure Sentinel.
Brought to you by Dr. Ware, Microsoft Office 365 Silver Partner, Charleston SC.