Designing an Adaptive Security Architecture for Protection From Advanced Attacks

Enterprises are overly dependent on blocking and prevention mechanisms that are decreasingly effective against advanced attacks. Comprehensive protection requires an adaptive protection process integrating predictive, preventive, detective and response capabilities.

Key Challenges

• Existing blocking and prevention capabilities are insufficient to protect against motivated, advanced attackers.
• Most organizations continue to overly invest in prevention-only strategies.
• Detective, preventive, response and predictive capabilities from vendors have been delivered in nonintegrated silos, increasing costs and decreasing their effectiveness.
• Information security doesn't have the continuous visibility it needs to detect advanced attacks.
• Because enterprise systems are under continuous attack and are continuously compromised, an ad hoc approach to "incident response" is the wrong mindset.

Recommendations

Information security architects:

• Shift your security mindset from "incident response" to "continuous response," wherein systems are assumed to be compromised and require continuous monitoring and remediation.
• Adopt an adaptive security architecture for protection from advanced threats using Gartner's 12 critical capabilities as the framework.
• Spend less on prevention; invest in detection, response and predictive capabilities.
• Favor context-aware network, endpoint and application security protection platforms from vendors that provide and integrate prediction, prevention, detection and response capabilities.
• Develop a security operations center that supports continuous monitoring and is responsible for the continuous threat protection process.
• Architect for comprehensive, continuous monitoring at all layers of the IT stack: network packets, flows, OS activities, content, user behaviors and application transactions.

Strategic Planning Assumptions

By 2020, 60% of enterprise information security budgets will be allocated to rapid detection and response approaches — up from less than 10% in 2014.

By 2020, 40% of enterprises will have established a security data warehouse — up from less than 5% in 2014.

By 2018, 80% of endpoint protection platforms will include user activity monitoring and forensic capabilities — up from less than 5% in 2013.

Introduction

This document was revised on 1 May 2014. The document you are viewing is the corrected version. For more information, see the Corrections page on gartner.com.

Most enterprise security protection efforts and products have focused primarily on blocking and prevention techniques (such as antivirus) as well as on policy-based controls (such as firewalls), to block threats (the upper-right quadrant of Figure 1). However, perfect prevention is impossible. Advanced targeted attacks are easily bypassing traditional firewalls and signature-based prevention mechanisms. All organizations should now assume that they are in a state of continuous compromise. However, organizations have deluded themselves into believing that 100% prevention is possible, and they have become overly reliant on blocking-based and signature-based mechanisms for protection. As a result, most enterprises have limited capabilities to detect and respond to breaches¹ (the bottom half of Figure 1) when they inevitably occur, resulting in longer "dwell times" and increased damage.

Figure 1. The Four Stages of an Adaptive Protection Architecture

Source: Gartner (February 2014)

In reality, going forward, improved prevention, detection, response and prediction capabilities are all needed to deal with all types of attacks, "advanced" or not (see Note 1). Furthermore, these should not be viewed as siloed capabilities; rather, they should work intelligently together as an integrated, adaptive system to constitute a complete protection process for advanced threats.

Analysis

To help enterprises design an architecture and select from among competing solutions for adaptive protection from advanced threats, we have developed an architecture composed of four high-level categories of competencies, with three drill-down capabilities in each category, for a total of 12 capabilities (described in more detail later in this research). It is necessary to focus on capabilities within each category to deliver comprehensive, adaptive protection from attacks.

• "Preventive" describes the set of policies, products and processes that is put in place to prevent a successful attack. The key goal of this category is to raise the bar for attackers by reducing their surface area for attack, and by blocking them and their attack methods before they impact the enterprise.
• "Detective" capabilities are designed to find attacks that have evaded the preventive category. The key goal of this category is to reduce the dwell time of threats and, thus, the potential damage they can cause. Detection capabilities are critical because the enterprise must assume that it is already compromised.
• "Retrospective" proficiencies are required to investigate and remediate issues discovered by detective activities (or by outside services), to provide forensic analysis and root cause analysis, and to recommend new preventive measure to avoid future incidents.
• "Predictive" capabilities enable the security organization to learn from external events via external monitoring of the hacker underground to proactively anticipate new attack types against the current state of systems and information that it is protecting, and to proactively prioritize and address exposures. This intelligence is then used to feed back into the preventive and detective capabilities, thus closing the loop on the entire process.

The adaptive protection architecture is a useful framework to help enterprises classify existing and potential security investments to ensure that there is a balanced approach to security investments. Rather than allowing the "hot" security startup of the day to define security investments, security organizations should evaluate their existing investments and competencies to determine where they are deficient. The adaptive protection architecture is also useful in classifying and evaluating vendors. Those that provide capabilities in multiple categories are more strategic than vendors that only fit in one category.

In an era of continuous compromise, enterprises need to shift from a mindset of "incident response" — wherein incidents are thought of as occasional, one-off events — to a mindset of continuous response — wherein attacks are relentless, hackers' ability to penetrate systems and information is never fully blocked, and systems must be assumed to be continuously compromised, and, thus, they must be continuously monitored (see Figure 2).

Figure 2. Continuous Monitoring Required for an Adaptive Protection Architecture

Source: Gartner (February 2014)

As shown in Figure 2, to enable a truly adaptive and risk-based response to advanced threats, the core of a next-generation security protection process will be continuous, pervasive monitoring and visibility that are constantly analyzed for indications of compromise. This will generate significant amounts of data. However, big data is only big noise unless appropriate analytics (supplemented with external sources of context, community and threat intelligence to improve accuracy) are used to distill it into actionable insight for the enterprise. The data can be analyzed using a variety of techniques, including heuristics, statistical analysis, inference modeling, machine learning, clustering analysis, entity link analysis and Bayesian modeling.

We believe that, going forward, all effective security protection platforms will include domain-specific embedded analytics as a core capability, in addition to traditional security information and event management (SIEM) systems. Enterprise monitoring should be pervasive and encompass as many layers of the IT stack as possible, including network activity, endpoints, system interactions, application transactions and user activity monitoring. This visibility must include enterprise-owned and employee-owned devices, and it must span enterprise data centers as well as the consumption of services from cloud-based providers.² The future of defense in-depth lies not only in layers of controls, but also in layers of monitoring and visibility (see Figure 3).

Figure 3. Continuous Monitoring of All Technology Layers

Source: Gartner (February 2014)

An enterprise's continuous monitoring of all entities and layers will generate a greater volume, velocity and variety of data than traditional SIEM systems can effectively monitor. This is one reason why Gartner research has established that big data analytics will be brought to next-generation security protection solutions, and also one of the reasons why, by 2020, 40% of enterprises will have established a "security data warehouse" for the storage of this monitoring data to support retrospective analysis. By storing and analyzing the data over time, as well as by incorporating context and including outside threat and community intelligence, patterns of "normal" can be established and data analytics can be used to identify when meaningful deviations from normal have occurred. As technologies supporting these capabilities become more mainstream, we believe that the adaptive protection architecture will also move into the mainstream as platform vendors that have numerous component pieces integrate the capabilities and provide an embedded analytics engine that is pretuned and ready to use out of the box.

Before we explore the 12 capabilities of the adaptive protection architecture, there are six key inputs that should be an integral part of the architecture and used throughout the process for security decision making (see Figure 4).

Figure 4. Policy, Context, Vulnerability Insight, Community Intelligence and Threat Intelligence Are Critical for Comprehensive Protection

Source: Gartner (February 2014)

Policy: Policies define and express the organization's requirements for system configuration, patching requirements, network connectivity, applications that are allowed to be executed, applications that are banned, anti-malware scanning frequency, sensitive data protection, what to do in the event of an outbreak and so on. These are typically derived from internal guidelines and external influences, such as regulatory requirements. Policies drive how enterprise security platforms will proactively prevent and reactively respond to advanced threats.

Context: Context-aware security (see Note 2) is the use of supplemental information to improve information security decision making at the time a decision is made, based on current conditions (for example, location, time of day, vulnerability state and so on). The use of context will be critical to identifying attacks that have bypassed traditional security protection mechanisms, and in helping to identify meaningful deviations from normal behavior without increasing the amount of false positives.

Community intelligence: To better protect against advanced threats, information should be aggregated, analyzed and shared across cloud-based communities that, ideally, have the ability to aggregate and analyze data for organizations in similar industries and geographic regions. This "crowdsourced" intelligence can then be shared to improve the overall protection capabilities of all participants. For example, community intelligence will help answer questions such as, "What are other enterprises like mine seeing? Have other people encountered this application/URL/IP address before? Has one of my peers developed a new way to detect an advanced threat and made this information available to others?" Thus, better communities will enable enterprises to share best practices, knowledge and techniques in a peering fashion. Larger communities will benefit from a network effect. Some communities will be self-forming, like FS-ISAC;³ some will be government-sponsored, such as the United States Computer Emergency Readiness Team (US-CERT); and others will be created by the security vendor, its partner ecosystem and users of its platform.⁴

Threat intelligence: The core of threat intelligence will be reputation feeds that provide insight into the trustworthiness of objects — for example, IP addresses, domains, URLs, files, applications and so on. However, advanced threat intelligence services will also provide enterprises with insight into how attackers and campaigns are organized and what specific targets they are attacking. In addition, these services will provide specific guidance on how enterprises can protect their systems and information from these attackers. Increasingly, threat intelligence is being delivered in machine-readable formats that are more easily and directly integratable into network, Web, email and endpoint security platforms that are designed to consume them.

Vulnerability insight: This information provides insight on vulnerabilities to devices, systems, applications and interfaces that the enterprise may have in use. In addition to known vulnerabilities, this insight should include visibility into unknown vulnerabilities that are present in an enterprise's custom and third-party applications. This can be accomplished by proactively testing these applications, libraries and interfaces for unknown vulnerabilities.

Vendor labs: Most security protection platform vendors provide information feeds that directly support their protection solutions — for example, signature updates as well as rule and pattern updates to provide protection from newly discovered threats.

To enable a comprehensive adaptive security protection architecture, we believe that 12 specific capabilities are necessary to augment our ability to block and prevent attacks, as well as detect and respond to attacks (see Figure 5).

Figure 5. 12 Critical Capabilities of Gartner's Adaptive Security Architecture

Source: Gartner (February 2014)

Below is a brief description of the 12 categories of capabilities, starting in the upper-right quadrant and moving clockwise. Note that the ordering does not imply importance; rather, all 12 capabilities should be considered equally important for comprehensive protection.

Harden and isolate systems: We believe the foundation of any information security protection architecture should start by reducing the surface area of attack by using a combination of techniques. These techniques limit a hacker's ability to reach systems, find vulnerabilities to target and get malware to execute. Traditional "default deny" (also referred to as "whitelisting") is a powerful capability and approach that falls into this category, whether at the network firewall (only communicate on this port/protocol) or the system application control level (only allow these applications to execute). Data encryption can be thought of as a form of whitelisting and hardening at the information level. Vulnerability and patch management approaches to identify and close vulnerabilities also map to this category. Emerging endpoint isolation and "sandboxing" techniques, which proactively limit the ability of a network/system/process/application to interact with others, are another example from this category.⁵

Divert attackers: Simply stated, techniques applied in this evolving category try to address the asymmetric advantages that hackers have in time. These techniques waste hackers' time by making it more difficult for them to locate legitimate systems and vulnerabilities to attack, hiding or obfuscating system interfaces and information through a variety of techniques (such as the creation of fake systems, vulnerabilities and information). For example, the Mykonos technology acquired by Juniper Networks creates the illusion of application layer vulnerabilities where none exist,⁶ thereby providing an active form of honeypots.⁷ Unisys Stealth hides networked systems,⁸ and CSG's Invotas solution implements a variety of diversion techniques.⁹ While security through obscurity is insufficient, it is appropriate to consider these capabilities in a layered, defense-in-depth protection strategy.

In addition to wasting hackers' time, these techniques can, with high assurance, quickly identify anyone trying to access fake systems, vulnerabilities and information as a hacker (since legitimate users would not be accessing these), and prevent them from causing damage. At the user interface layer, newer vendors, such as Shape Security,¹⁰ harden applications at the user interface layer to protect against automated attacks.

Prevent incidents: This category maps to well-established approaches to prevent hackers from gaining unauthorized access to systems; it includes traditional "signature based" anti-malware scanning as well as network and host-based intrusion prevention systems. "Behavioral signatures" may also be used at different layers here — for example, to prevent systems from communicating with known command-and-control centers by using threat intelligence from third-party reputation service feeds and integrating it into network, gateway or host-based controls (or within a host, thereby preventing one process from injecting itself into the memory space of another).

Detect incidents: Some attacks will inevitably bypass traditional blocking and prevention mechanisms, in which case it is key to detect the intrusion in as short a time as possible to minimize the hacker's ability to inflict damage or exfiltrate sensitive information. A variety of techniques may be used here (see Note 3), but most rely on the analysis of data gathered by continuous monitoring at the core of the adaptive protection architecture, by detecting anomalies from normal patterns of network or endpoint behavior, by detecting outbound connections to known bad entities, or by detecting sequences of events and behavioral signatures as potential indictors of compromise.

The continuous and pervasive monitoring at the heart of Figure 5 becomes critical to perform analytics on what is currently being observed versus what has been normal in the past so that the security operations analyst can identify anomalies. Going forward, the development of a continuous security operations center and skilled security operations analysts will become critical competencies for enterprises.

Confirm and prioritize risk: Once a potential incident has been detected, it needs to be confirmed by correlating indicators of compromise across different entities — for example, comparing what a network-based threat detection system sees in a sandboxed environment to what is being observed on actual endpoints in terms of processes, behaviors, registry entries and so on. This ability to share intelligence across networks and endpoints is one of the primary reasons cited by FireEye in its recent acquisition of Mandiant.¹¹ Based on internal and external context — such as the user, his or her role, the sensitivity of the information being handled and the business value of the asset — this issue should be prioritized by the risk to the enterprise, and be visually presented so that the security operations analyst can focus on the highest-risk priority issues first.

Contain incidents: Once an incident has been identified, confirmed and prioritized, this category works to contain the threat by isolating the compromised system or account from accessing other systems. Common containment capabilities are, for example, endpoint containerization, account lockout, network-level isolation, killing a system process, and immediately preventing others from executing the same malware or accessing the same compromised content.

Investigate/forensics: Once the compromised systems or accounts have been contained, the root cause and full scope of the breach should be determined using retrospective analysis of what exactly happened, using the data gathered from the ongoing and continuous monitoring at the core of Figure 5. How did the hacker gain a foothold? Was an unknown or unpatched vulnerability exploited? What file or executable contained the attack? How many systems were impacted? What specifically was exfiltrated? In some cases, enterprises may want to know more about the origin and motivation of the hackers — for example, Was this a nation-state-sponsored attack? If so, which nation? This category requires detailed historical monitoring information for the security analyst to answer these detailed questions. Network flow data alone may be insufficient for a complete investigation. More advanced security operations centers use full packet capture at the network (and the equivalent at the endpoint, in terms of system activity monitoring), along with associated advanced analytics tools, to answer these types of questions. Likewise, as new signatures/rules/patterns are delivered from the vendor's labs and research capabilities, they should also be run against historical data to see if the enterprise has already been targeted with this attack, and the attack has remained previously undetected.

Design/model change: To prevent new attacks or reinfection of systems, it is likely that changes to policies or controls will be needed — for example, vulnerabilities closed, network ports closed, signatures updated, system configurations updated, user permissions modified, user training changed or information protection options strengthened (such as encryption). More advanced platforms should be capable of automatically generating new signatures/rules/patterns to address newly discovered advanced attacks — in essence, providing a "custom defense." However, before these are implemented, the change should be modeled against the historical data that has been gathered from the continuous monitoring to proactively test for false positives and false negatives.

Remediate/make change: Once modeled and determined to be effective, the change must be implemented. Some responses can be automated using emerging security orchestration systems, and policy changes can be pushed to security policy enforcement points, such as firewalls, intrusion prevention systems (IPSs), application control or anti-malware systems. For example, there are emerging security response orchestration solutions that are designed to automate and orchestrate this process.¹² However, at this early stage, many enterprises still prefer that security operations specialists, network security specialists or endpoint support staff members implement the change, rather than automated systems.

Baseline systems: Changes will be continually made to systems; new systems (such as mobile devices and the use of cloud-based services) will be continually introduced; user accounts will constantly come and go; new vulnerabilities will be disclosed; new applications will be deployed; and ongoing adaptations to new threats will be made. Thus, there must be a continuous rebaselining and discovery of end-user devices, back-end systems, cloud services, identities, vulnerabilities, relationships and typical interactions.

Predict attacks: This category is emerging and growing in importance. Based on reconnaissance of hacker attention, hacker marketplaces and bulletin boards; on vertical industry interest; and on the type and sensitivity of the data being protected, this category is designed to proactively anticipate future attacks and targets so that enterprises can adjust their security protection strategies to compensate. For example, based on intelligence gathered that indicates a likely attack on a specific application or OS (see Note 4), the enterprise could proactively implement application firewalling protection, strengthen authentication requirements or proactively block certain types of access.¹³

Proactive exposure analysis: With the latest intelligence gathered internally and externally, exposure and risk to enterprise assets must be continually assessed against predicted and anticipated risks, and adjustments to enterprise policies or controls may be needed. For example, when consumption of new cloud-based services is discovered, what risk does this represent?¹⁴ Are compensating controls, such as data encryption, needed? The same is true for new applications that are discovered, whether they are enterprise applications or applications on mobile devices: What risk do these represent? Have they been scanned for known and unknown vulnerabilities? Are compensating controls, such as application firewalls or endpoint containment, needed?

Capabilities Must Work Together as a System

The end result should not be 12 silos of disparate information security solutions. The end goal should be that these different capabilities integrate and share information to build a security protection system that is more adaptive and intelligent overall. For example, while the enterprise may not have had a "signature" to prevent a breach initially, after the attack is discovered, the enterprise can use the knowledge gained by a forensic analysis of the attack to block further infections, in essence developing a "custom defense" against the attack. Thus, the notion that "signatures are dead" is misguided hyperbole. Signature-based prevention techniques still play a useful role in the process, even if the "signature" to block the attack from spreading comes after the initial breach. In another example, a network-based advanced threat detection appliance can exchange indicators of compromise with endpoints to confirm whether an attack has taken hold on enterprise systems.¹¹ Thus, the adaptive protection architecture works throughout the life cycle of an attack (see Figure 6).

Figure 6. Mapping the Adaptive Protection Process to the Life Cycle of an Attack

Source: Gartner (February 2014)

Security intelligence emerges from this continuous process as the categories of capabilities and different layers of security controls exchange intelligence, creating a need for a new generation of Intelligence-Aware Security Controls (IASC). Like integrating threads of fiber in a rope, the integration of the capabilities, the exchange of intelligence between them, and the exchange of intelligence to and from the community and threat intelligence providers deliver overall greater protection.

Complete protection requires prevention, detection, retrospective analysis and predictive capabilities. More capable security protection platforms will include competencies in more stages and more of the specific drill-down capabilities in each stage. For example, the next generation of network security platforms should include firewalling, intrusion prevention capabilities and detection capabilities, such as content analysis capabilities.¹⁵

Furthermore, there is an opportunity for vendors that span different layers in the stack to provide a more integrated offering across different layers of the IT stack. For example, a vendor that has network-based protection and endpoint protection capabilities may link these for improved overall protection. Where a vendor doesn't directly have capabilities in an area, it should partner to improve the protection capabilities of its offerings.

The ability to integrate with external context and intelligence feeds, as shown in Figure 4, is also a critical differentiator. For example, what types of context — location, time of day, device, reputation and so on — can the vendor understand and incorporate into its security decision making? Does the vendor support and nurture a robust cloud-based community of its customers for the exchange of community security intelligence? What types of reputation feeds can the platform support for improved security protection — for example, taking into consideration IP, URL, device, file and user reputation — in the security decision-making process?

Finally, we believe that leading next-generation security platforms should provide risk-prioritized actionable insight derived from embedded domain-specific analytics capabilities within the platform. The built-in analytics capabilities will work against the data gathered from the continuous monitoring at the center of these platforms to deliver the actionable insight at the top of the pyramid in Figure 2.

The goal is not to replace traditional SIEM systems, but rather to provide high-assurance, domain-specific, risk-prioritized actionable insight into threats, helping enterprises to focus their security operations response processes on the threats and events that represent the most risk to them. SIEM systems will still be needed to support near-real-time detection of threats across different layers of monitoring data, and, rather than blindly consuming all events, these systems will consume the prioritized, domain-specific intelligence produced by the next generation of security protection platforms, thus providing more effective SIEM results as well.

Source: Gartner Research G00259490, Neil MacDonald, Peter Firstbrook, 12 February 2014