-This paper outlines the need for and the development of an Incident Management Ontology. The Incident Management Ontology is derived from an Incident Management Meta-Model. We describe the shortcomings of the Incident Management Meta-Model and how the Incident Management Ontology addresses these shortcomings. The development of the Incident Management Ontology is outlined and the need for such an ontology is discussed. Related work is described and the Incident Management Ontology's potential uses and applications are presented.
When the JASON1 Program within MITRE looked at the scientific community for ways to make cybersecurity “more scientific”, their very first conclusion was that the security community needed “a common language and a set of basic concepts about which the security community can develop a shared understanding” [1], or in other words, a Cybersecurity ontology. The work described in this report is part of an ongoing effort within CERT® to build such an ontology for incident management.
We believe that such formal models are the best way for the community to evolve towards a “science of cybersecurity”, and that our incident management ontology can play a crucial role in improving incident management. The ontology’s purpose is to create a common language for describing the processes and functions associated with CSIRTs. We intend to use the ontology to analyze existing CSIRTs, to define a standard set of processes and services that should be offered by CSIRT teams, to formalize roles and responsibilities, and to build an ontology based competency model for the knowledge, skills, and abilities required of team members.
This paper describes the evolution of our work on characterizing incident security teams from a natural-language text document to a formal ontology and analyzes the benefits that accrued in the process. When creating our ontology, we 1 “JASON is an independent scientific advisory group that provides consulting services to the U.S. government on matters of defense science and technology. [In 2010] JASON was asked by the Department of Defense to examine the theory and practice of cyber-security, and to evaluate whether there are underlying fundamental principles that would make it possible to adopt a more scientific approach.” (http://fas.org/irp/agency/dod/jason/) chose to use the W3C Ontology Web Language - OWL2 due to its formalism and increasing use in the Semantic Web community. We feel this work may be a useful case study for others who are thinking about formalizing their own information security knowledge.
In previous work [2], we aggregated a wide variety of incident management process models such as ISO 27002 [3] and NIST 800-61 [4]. From those sources we abstracted a generalized meta-model that captured the essential processes involved in incident management.
This meta-model was at the heart of what we previously called an Incident Management Body of Knowledge (IMBOK). It broke incident management activities into 18 high-level tasks organized by the incident management life cycle phases as Prepare, Protect, and Respond. It also included five nonprocedural, crosscutting capabilities that constrain all the other tasks. The following outlines the phases and tasks and 5 crosscuts of the IMBOK: A. The phases and tasks 2 OWL is based upon description logics. OWL supports those users who want the maximum expressiveness while retaining computational completeness (all conclusions are guaranteed to be computable) and decidability (all computations will finish in finite time). (http://www.w3.org/TR/owl-features/) • • • • • • •
Collect and preserve evidence Restore and validate the system Perform a postmortem review of incident management actions Integrate lessons learned with problem management process Analyze incident, including artifacts, causes, and correlations
Determine and remove the cause of the incident B. The 5 crosscuts 1) Manage information 2) Properly handle collected evidence following best practices 3) Manage the incident management team 4) Communicate incidents 5) Track and document incidents from initial detection through final resolution C. Drawbacks to the Incident Management Meta-Model Although the Incident Management Meta-Model provides a considerable simplification and consolidation of prior knowledge, it suffers from a number of drawbacks due to its knowledge representation formalism: • The use of imperative verb forms expressing infinitive constructions means that each task is only partially represented, because the subject is implicit. This obfuscates, for example, the fact that some of the tasks (e.g. managing the team) are carried out by the team's managers, not by the incident responders. • In general, the use of natural language makes machine processing of this knowledge representation difficult. • In particular, there is no easy way to use this representation to perform modeling and simulation, nor to build applications on top of it. • To keep the process model manageable, concepts have been abstracted to an unusable level, with no graceful way to expand them into a more detailed form. There is no way within this system, for example, to say what is meant by "defenses" in "improve defenses". • Apart from including a glossary, this representation does not facilitate the use of a standardized vocabulary. • Also to keep the process model manageable, related concepts have been combined, as in "restore and validate the system". • Despite its relative compactness, this representation violates the "7 plus or minus 2" law [5] and is hard for users to take in at a glance and internalize.
natural-language format of the body of knowledge, and instead building a formal ontology using OWL.
An ontology is simply a set of shared, precisely-defined concepts in a given domain, along with the relationships among those concepts. OWL (the Web Ontology Language) is a W3C recommendation that builds on earlier languages from DARPA and elsewhere [6], is a key component of the Semantic Web [7], and is currently the leading knowledge representation and reasoning language in computer science. OWL is descended from earlier attempts at usable knowledge representation systems such as expert systems, logical programming languages, frame-based reasoning systems, modal logic, KL-One [8], entity-relationship modeling, and the like [9]. Description Logics emerged as a flexible yet powerful knowledge representation tool as the relationships among these approaches were better understood and new ways to engineer logics and reasoning systems were discovered. Description Logics have been used projects ranging from the International Catalogue of Diseases [10] to Google's Knowledge Graph [11].
To build our IM ontology, we decomposed the 18 highlevel tasks in the IMBOK meta-model into component concepts and their respective relationships.The concepts, also known as classes in the Description Logic community, are organized into a strict hierarchy of subclasses. The incident management tasks are composed of relationships among those classes. This separation of classes from relationships is the key to most modern knowledge formalisms, from KL-One [8] to OWL [12].
The only relationships inherent in the Description Logic on which OWL is built are binary relationships consisting of two concepts (or objects) and a relationship between them. However, many of the relationships we want to model in incident management are "n-ary" relationships among more than just two objects. For example, training requires a relationship among at least three objects: the training itself, a trainer, and a trainee.
There are a number of ways to handle this situation in OWL; for the IM ontology we used one of the techniques recommended by the W3C [13]. This technique consists of creating a new class that holds the relationships among the training concepts.
This requires a slight adjustment to our ways of thinking about relationships. To illustrate, the original meta-model tasks
(IM leaders) Develop trusted relationships with external experts.
(trainers) Provide staff with appropriate education and training. become developing external relationships: involves external groups produces trusted relationships is performed by IM leaders staff training:
is provided by either external or internal trainers
is provided to IM personnel
Once the reified relationships are in place, it becomes straightforward to enhance them with additional information. In full, these two classes actually are as follows in the ontology: developing external relationships: belongs to the prepare process involves external groups produces trusted relationships is subject to the incident management crosscuts
is performed by IM leaders staff training:
is provided by either external or internal trainers is provided to IM personnel is a training service is part of the prepare process is subject to the incident management crosscuts
The table In Appendix B gives a simple summary of the relationships in the ontology.
Figure 1 shows a screenshot of the IM ontology being edited in Protégé [14], the ontology development tool from Stanford that is widely used in the community. The display contains five panes giving five views of the ontology. The upper left pane shows the class hierarchy.
The two most important classes are "activities" and "crosscuts". The activities are simply the tasks carried out by the incident management staff, while "crosscuts" or "principles" as Beebe and Clark call them [15] are pervasive constraints on the activities. In addition to those main classes, we needed eight auxiliary classes to describe the activities in full: incident components, IT components, knowledge assets, life cycle phases, organizational groups, quality standards, relationships, and team resources. These classes were identified using traditional ontology-mining techniques: we started with the terms in the meta-model, then clustered them and introduced class hierarchies based on our knowledge of the domain.
The top two right-hand panes of the display show additional information about the selected class in the class hierarchy, in this case "incident response services". The top pane shows the usage of the selected class, while the second pane shows information about the class in terms of its subclasses, its superclasses, its members, any equivalent classes, and so forth.
The pane at the lower left of the screen shows the hierarchy of relationships, called "object properties" in OWL. The "permeates" relationship has been selected. The lowest pane on the right describes that relationship in the ontology, showing that its domain is "CSIRT managers" and its range is "team resources", capturing the fact that
CSIRT team managers acquire the team's resources.
C. Overcoming the Drawbacks
We believe that this formal IM ontology solves the problems noted in Section 2 for the IMBOK meta-model.
• The use of classes and relationships ensures that the
knowledge is represented completely. • This representation is machine-processable; Figure 2 shows a simple graphic automatically generated from the IM ontology using the OntoGraf tool [16] with a
GraphViz post-processing script. • The use of Description Logic (DL) ontologies for modeling and for constructing applications is well understood [9]. • The use of a strict class hierarchy gives us a userfriendly way to talk about concepts at any needed level of abstraction without complicating the IM ontology as a whole: we can talk about "security training", or "training", or "proactive services", or "incident management services", and the reasoning system will infer properties and type relationships as needed. • The use of OWL annotations to capture definitions
makes the IM ontology usable as a dictionary. • Because of the class hierarchy and the formality of the system, there is no pressure to collapse concepts to keep the document small. • Finally, the separation of entities from relationships reduces the complexity of the representation, and makes the structure of the IM ontology easier to absorb.
Figure 2 gives a high-level breakdown of the incident management activities. The "service delivery" activities are the most important, and Figure 2 expands that class to a further level of detail. Figure 3 shows a close-up of the root cause analysis environment, showing that it is performed by incident management personnel, that its goal is to explain root causes, that it is an incident analysis service, and so forth. D. Other Benefits
In addition to solving the difficulties we had with the MetaModel, moving to a formal ontology had several other advantages.
1) Very flexible typing
We quickly grew fond of the ability to create new types simply by specifying the necessary and sufficient conditions for membership in the type. Earlier we had used a multidimensional organization system called facet maps [17] to achieve multiple categorizations for the Meta-Model, but class expressions are much more lightweight and flexible.
They are like a very disciplined tagging system. To cite just one example, we realized at some point that although we want to retain the classification of activities by the life-cycle phase in which they are used (prepare, protect, detect, respond, etc.), there is no need to build the life-cycle phases into the class hierarchy. Instead we simply assert a "belongs-to" relationship between an activity and a life-cycle phase. Then we can define a "protect-activities" class where the membership condition is "an activity that belongs-to the protect phase" and the reasoner will automatically compute the members of the class.
2) More powerful Modeling
The n-ary relations that use binary relations to “reify” relations among individuals turned out to be a very effective method for packaging up domain knowledge in a taxonomic hierarchy. When it seemed clear that the different types of incident analysis were characterized by the goal of their analysis, it was trivial to add "explains" and "explained-by" relationships.
3) Improved knowledge visualization
A shortcoming of our Incident Management Meta-Model was the absence of a satisfactory visualization. After converting the Meta-Model into a formal ontology, we used OntoGraf [16] to export files in the GraphViz DOT format [18]. DOT is a text-based format that allows for customizable graphics.
E. Individuals
The real power of Description Logic ontology comes when an ontology is populated by individuals and reasoning is enabled. “Reasoning” is a key-functionality of semantic technologies and allows automatic inferences to be made using the rules and classes described by the ontology. The ability of OWL to be used at internet scale comes from the highly optimized and logically precise handling of both terminological, or taxonomic, knowledge in what the Description Logic community calls the TBox, and the contingent assertional knowledge about individuals in what the community calls the ABox [9].
We have not yet formally extended the Incident Management Ontology to real world individuals, but Figure 4 shows an example using two fictitious individual CSIRTs in the ontology. The Acme team, focused only on incident response, provides monitoring, incident detection, incident reporting, and incident analysis
Figure 4 - A Comparison of Two Fictitious Incident Management Teams. Note that “has individual” denotes the membership of individuals in classes. Thus Acme is an individual member of the class of incident reporting services.
IV. RELATED RESEARCH
The seminal paper Formalizing Information Security Knowledge by Fenz et al. [19] presents the rationale for capturing information security best practices in an OWL ontology. Though it served as an influence for our ontology, the work addresses information security in general while our work focuses on incident management.
There have been many proposals for standardized incident handling process models; for a summary of the models that were used for our meta-model, see [2]. Although they incorporated much collective wisdom, none of them were based on a formal knowledge representation. Like our metamodel, the forensic framework of Beebe and Clark [15] aimed to assimilate existing practice into a comprehensive framework. The distributed, loosely-coupled incident response model of Millar, Osorno, and Reger [20] is a deeply-reasoned attempt to analyze and improve upon existing incident management practices based on scientific theory and simulation, but is not based on a formal ontology.
Furthermore, we found that many proposed ontologies that exist fail to capture all of the important relationships between members of organizations and the organizations themselves.
These representations arise from an internal focus of an organization who has been victim to attack, and many ignore the roles and relationships between a CSIRT and incidents that occur.
Magklaras and Furnell [21] observe that incidents occur through misuse by individuals, but do not propose a formalized ontology of a taxonomy including this humanmisuse concept. Classifications of individuals are made more distinguished based on behavior (e.g. accidental or intentional), and possible consequences of misuse correlated to these actions.
Wang and Guo’s [22] research in developing OVM (Ontology for Vulnerability Management) identifies individuals responsible for attacks, but the relationships amongst these individuals is not made clear. The formalizations within their work capture knowledge sufficient to answer questions about the assets targeted in an incident and mechanisms by which an incident takes place. While organization and individuals are clear in this work, further subdivisions of organizations and groups of individuals are not. No concept of trust appears in the ontology’s class hierarchy, making the risk of agents difficult to reason about.
Chiang [23] proposed mapping the IT Security EBK [24] and ISO/IEC 27001 [25] standard to an incident ontology. The construct is similar to OVM, but has the benefit of subdivision of roles amongst individuals and groups. Subdivisions, however, are limited and the ontology will require additional, higher-level concepts to subsume various sibling classes of the hierarchy.
The most complete formalization framework in security that gathers all necessary information to incident management might be Ekelhart’s [26] move from simple security taxonomy to ontology. This work acknowledges the different threats and means for attacks, along with measurable reductions when safeguards are introduced. Even relationships amongst individuals in an organization and the roles they take are represented clearly. However, this research does not model subdivisions of an organization and the roles that multiple organizations can have (both within and in relation to one another). Different subdivisions of service types and measures of trust are not represented.
V. NEXT STEPS
Future work on the Incident Management Ontology will focus on evaluting the ontology and using it to categorize incident response organizations. This work names CSIRT processes but does not yet describe them in full detail. Future work may include using existing standards, such as the Process Specific Language[27], to model the process flows for each service offered by a CSIRT in greater detail. We plan to evaluate the ontology by using it to analyze the processes performed by and services offered by incident response teams.
A hypothesis we would like to test is whether there is a difference between the functions of CSIRTs and Coordination Centers. We are collecting data on both types of organizations and plan to analyze it using the ontology. We also plan to improve the ontology by adding axioms, more defined classes, and taking more advantage of reasoning capabilities.
VI. ACKNOWLEDGEMENTS This material is based upon work funded and supported by the Department of Defense under Contract No. FA8721-05-C0003 with Carnegie Mellon University for the operation of the Software Engineering Institute, a federally funded research and development center. This material has been approved for public release and unlimited distribution. Carnegie Mellon® and CERT® are registered Mellon University. DM-0001433 marks of
Carnegie NO WARRANTY. THIS CARNEGIE MELLON UNIVERSITY AND SOFTWARE ENGINEERING INSTITUTE MATERIAL IS FURNISHED ON AN “AS-IS” BASIS.
CARNEGIE MELLON UNIVERSITY MAKES NO WARRANTIES OF ANY KIND, EITHER EXPRESSED OR IMPLIED, AS TO ANY MATTER INCLUDING, BUT NOT LIMITED TO, WARRANTY OF FITNESS FOR PURPOSE OR MERCHANTABILITY, EXCLUSIVITY, OR RESULTS OBTAINED FROM USE OF THE MATERIAL. CARNEGIE MELLON UNIVERSITY DOES NOT MAKE ANY WARRANTY OF ANY KIND WITH RESPECT TO FREEDOM FROM PATENT,
TRADEMARK, OR COPYRIGHT INFRINGEMENT. APPENDIX A: AN OVERVIEW OF THE ACTIVITIES CLASS IN THE ONTOLOGY • • • • • • • • • • • • • • • • • • •
Incident management (IM) leaders develop trusted relationships with external groups Both internal and external trainers provide training to IM personnel internal trainers provide awareness training to partners IM leaders develop governance artifacts IM leaders perform management functions on IM personnel IM leaders develop planning artifacts IM personnel provide vulnerability remediation to constituents IM personnel provide incident detection to constituents IM personnel provide incident communication to constituents IM personnel provide defense hardening to constituents IM personnel perform triage incident data collectors perform incident data collection IT personnel restore IT components IT personnel validate IT components IM personnel coordinate analyzing lessons learned IM incident handlers perform incident analysis IM personnel perform incident resolution IM leaders perform management functions IM personnel perform incident tracking APPENDIX B: THE CLASS HIERARCHY OF THE IM ONTOLOGY This appendix contains the class hierarchy in the Incident Management Ontology.
activities: functions performed by a CSIRT developing-governance: establishing the operational guidelines for an organization developing-plans: establishing and maintaining the business and operational plans for an organization developing-policies: establishing and maintaining the policies that guide the organizational activities developing-procedures: establishing and maintaining implementations of organizational policies developing-processes: establishing and maintaining organizational processes develop-data-collection-processes: establishing logs and monitoring to provide insight into incidents developing-relationships: identifying and communicating with essential business partners developing-external-relationships: developing relationships with external parties developing-internal-relationships: developing relationships with internal parties prepare-activities: activities that are typically carried out during the prepare phase of the incident life cycle process-improvement: activity whose goal is to improve the efficiency, reproducibility, reliability, or other quality attribute of business processes integrating-lessons-learned: feeding the results of a postmortem review into the organization's problemsolving process postmortem-review: an examination of an event to discover factors that affected the quality of the handling of the event measuring-performance: collecting metrics that assess the quality of a process for process improvement purposes protect-activities: activities that are typically carried out during the protect phase of the incident life cycle respond-activities: activities that are typically carried out during the respond phase of the incident life cycle service-delivery: the activity of providing a service to a constituent defense-hardening-service: assisting with improving the security defenses of a constituent improve-defenses: hardening defenses by improving the security controls in place remediation-service: hardening defenses by removing known vulnerabilities and risks risk-assessment-service: hardening defenses by identifying threats vulnerability-assessment-service: hardening defenses by identifying vulnerabilities incident-response-service: providing assistance in responding to and recovering from incidents evidence-collection-service: gathering and maintaining information concerning an event diagnostic-data-collection-service: incident-data-collection to support diagnosis and restoration activities forensics-data-collection-service: incident-data-collection to support legal activities incident-analysis-services: using collected data to uncover the causes and time-line of an event artifact-analysis-service: incident analysis applied to artifacts incident-analysis-service: general incident analysis root-cause-analysis-service: incident analysis with the goal of determining the root cause of an event vulnerability-analysis-service: incident analysis applied to the vulnerability that enabled an event incident-detection-service: proactive steps to ensure events and incidents are discovered and reported as soon as possible incident-recovery-service: reactive activities with the goal of restoring an affected system to the state before an event system-restoration-service: restoring an affected system to the state before an event system-validation-service: verifying that an affected system has been restored incident-reporting-service: communicating information about an event or incident in accordance with an incident reporting policy monitoring-service: maintaining an automated infrastructure to detect events and report incidents training-service: a proactive service to ensure that stakeholders have the knowledge, skills, and abilities they need constituent-training-service: training for constituents that helps them protect their infrastructure staff-training-service: training for staff that helps them perform their jobs team-training-coordination: ensuring adequate training for staff sustain-activities: activities whose goal is to prevent the CSIRT's posture from declining over time crosscuts: constraints or principles that apply to activities incident-communication: communicating information about the effects of a incident to staff and constituents incident-coordination: ensuring that all IM stakeholders are with a shared plan incident-data-collection: collection of data relevant to an incident incident-documentation: documenting the results of incident-analysis incident-problem-solving: using generic or specialized methods in an orderly manner to find solutions to problems incident-resolution: an action taken to repair the root cause of an incident or to implement a workaround incident-tracking: managing and maintaining a database of information on incidents and constituents incident-components: the various elements that constitute the conceptual model of an event artifacts: any entities left behind after an incident takes place; for example, malicious code or logfiles events: any occurrences that may have negative security consequences incidents: events that have been confirmed to have negative security consequences root-causes: the earliest occurrence in the causal chain leading to an incident vulnerabilities: the weaknesses in the system that were exploited by an incident IT-components: the various elements that constitute the conceptual model of an IT system information-system: collection of technical and human resources that provide storage, computing, and distribution for enterprise information network: collection of host computers together with the sub-network or inter-network through which they can exchange data security-tools: hardware and software that improves the security of the information-system in which they are installed incident-detection-tools: security-tools that perform incident-detection av-systems: incident-detection-tools that work by analyzing virus signatures ids-systems: incident-detection-tools that work by analyzing activity on the network network-monitors: security-tools that work by observing network activity knowledge-assets: the various types of documents that constitute the intellectual capital of the organization governance-artifacts: documents that are used in the process of governing policies: abstract documents that express decisions made by management about the running of the organization procedures: concrete documents that implement policies processes: workflows that implement policies and procedures incident-reports: documents that inform the CSIRT about events and incidents incident-tracking-documents: case management documents that trace the progress of an event through the incident-handling process incident-assignments: tagging of incidents with the names of IM-personnel responsible for handling them incident-categorization: tagging of incidents with the classification into which they fall information: general documents that do not fall in any other category lessons-learned: documents that capture the results of analyzing-lessons-learned other-knowledge-assets: any information not included in other categories planning-artifacts: abstract documents that prepare IM-personnel for incident response incident-response-plans: planning-artifacts that reflect decisions made about incident-response within the organization incident-response-strategies: technical documents that guide IM-personnel in responding to incidents training-materials: documents that are used to provide training life-cycle-phase: the temporal periods into which incident response is divided prepare-phase: educating personnel and providing them with the tools needed to perform their jobs protect-phase: applying controls and otherwise hardening the infrastructure to resist attack respond-phase: detecting, analyzing, and recovering from incidents sustain-phase: ensuring that the capability of the CSIRT does not degrade over time organizational-groups: stakeholders in the incident management process external-groups: stakeholders not within the administrative boundaries of the organization external-csirts: incident management teams outside the boundaries of the organization external-trainers: educational personnel outside the organization law-enforcement-agencies: external groups performing law enforcement functions other-external-groups: any other external group partners: groups or sets of individuals with close relationships to the organization constituents: the groups or sets of individuals for whom incident management is being performed staff: stakeholders contained within administrative boundaries of the organization
IM-personnel: groups or sets of individuals tasked with performing incident management
IM-incident-handlers: individuals responsible for responding to and recovering from incidents IM-forensics-analyst: an IM-incident-analyst specializing in analysis for legal purposes IM-incident-analyst: an IM-incident-handler specializing in the analysis of incidentcomponents
IM-malware-analyst: an IM-incident-analyst specializing in reverse engineering
IM-leaders: individuals responsible for leading the incident management personnel incident-data-collectors: individuals responsible for collecting data about incidents diagnostic-data-collectors: incident-data-collectors that collect data for diagnostic purposes forensic-data-collectors: incident-data-collectors that collect data for forensic purposes internal-trainers: educational personnel within the organization IT-personnel: members of the it staff that carry out security functions such as infrastructure hardening management: individuals responsible for governing line-management: managers at the low end of the chain of command mid-level-management: managers in the middle of the chain of command senior-management: managers at the high end of the chain of command risk-assessors: individuals responsible for assessing risks to the organization vulnerability-assessors: individuals responsible for identifying vulnerabilities in the organization's infrastructure quality-standards: normative requirements for ensuring the high quality of the CSIRT's activities document-management-standards: standards that constrain the way information is handled within the organization appropriately-dissemination-standards: standards that govern the provision of information to the appropriate audiences confidentiality-preserving-standards: standards that govern how information is withheld from inappropriate audiences forensic-standards: quality standards that ensure the admissibility of the analysis in a court of law preserving-chain-of-custody: documenting that there has been no opportunity for forensic evidence to be tampered with other-quality-standards: quality standards not included in other categories relationships: connections between individuals or groups of individuals trusted-relationship: relationships among entities that are willing to share confidential data untrusted-relationship: relationships among entities that are willing to share confidential data team-resources: anything needed for the CSIRT activities or the operations of IM-personnel funding: financial resources necessary for the operations of IM-personnel IT-infrastructure: information security assets necessary for the operations of IM-personnel staffing: human resources necessary to ensure the operations of IM-personnel