Systematic Risk Assessment of Cloud Computing Systems using a
Combined Model-based Approach
Nazila Gol Mohammadi
1
, Ludger Goeke
1
, Maritta Heisel
1
and Mike Surridge
2
1
Working Group Software Engineering, University of Duisburg-Essen, Oststr. 99, Duisburg, Germany
2
IT Innovation, Southampton, U.K.
Keywords:
Context Analysis, Risk Assessment, Threat and Control Identification, Cloud Computing Systems.
Abstract:
Data protection and a proper risk assessment are success factors for providing high-quality cloud computing
systems. Currently, the identification of the relevant context and possible threats and controls requires high
expertise in the security engineering domain. However, consideration of experts’ opinions during the devel-
opment life-cycle often lacks a systematic approach. This may result in overlooking of relevant assets or
missing relevant domain knowledge, etc. Our aim is to bring context analysis and risk assessment together
in a systematic way. In this paper, we propose a systematic, tool-assisted, and model-based methodology to
scope the context and risk assessment for a specific cloud system. Our methodology consists of two parts:
First, we enhance the initial context analysis necessary for defining the scope for risk assessment, and second
we identify relevant threats and controls during design- and deployment-time. Using the context model, and
design-time system model, we further refine the gathered information into a deployment model. All steps of
our methodology are tool supported and in a semi-automatic manner.
1 INTRODUCTION
A proper information security risk management is a
key success factor for providing secure cloud com-
puting systems. Recently, a significant number of se-
curity incidents have been reported. Such incidents
can lead to strong consequences for cloud service
providers not only financially, but also in terms of rep-
utation loss.
The management of information security risks re-
quires activities to identify and control information
security risks for the assets of an organization. These
assets include, among others, processes and informa-
tion/data that are essential for providing the business
of an organization.
The establishment of a context for the information
security risk management is an important activity as
the context defines which assets are relevant for the
risk management. Because of the complex architec-
ture of cloud computing systems, the number of par-
ticipating stakeholders and relevant regulations, the
context establishment is a difficult task that carries the
danger that critical information is overlooked.
Subsequent to context creation, risk assessment
is a crucial task. The risk assessment identifies the
risks that assets may be compromised due to an in-
formation security incident. To this end, the threats
to the assets have to be identified. The large number
of threats for different types of assets and the rapidly
changing threat landscape make the threat identifica-
tion a difficult task, and some threats can easily be
overlooked. Moreover, dependencies between assets
give rise to potential attack paths (using sequences of
threats to individual assets), and also secondary effect
cascades. Either can also be easily missed, leading to
errors in the assessment of risks from the identified
threats.
Finally, although standardized approaches require
the context for risk management to be established
first, in practice the use of third party software, ser-
vices, and cloud resources means that the context for
the risk analysis is often determined only just before
a system is deployed in the cloud. Therefore, we aim
to bring the context analysis and a risk assessment
approach together to provide systematic guidance in
identifying the relevant context information and per-
forming threat and control identification as early as
possible, i.e. during requirements engineering and
design-time.
In this paper, we present a methodology that al-
lows to systematically identify relevant context infor-
mation and after that the identification of threats and
Mohammadi, N., Goeke, L., Heisel, M. and Surridge, M.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach.
DOI: 10.5220/0009342700530066
In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 2, pages 53-66
ISBN: 978-989-758-423-7
Copyright
c
2020 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
53
controls. Our methodology follows a model-based
approach. This ensures consistency and traceabil-
ity, and provides unambiguous representations of risk
factors and assumptions that can be used to facili-
tate communication between stakeholders. We pro-
vide tools supporting the use of each model, and these
tools guide risk analysts through the different steps of
our methodology and supports their execution, thus
limiting the manual effort to apply the methodology.
Furthermore, one of the tools provides a possibility to
verify the model which helps to detect errors in the
application of the methodology.
For the context establishment, we provide a graph-
ical context pattern with an underlying context model
that defines types of information that are relevant for
the context of cloud computing systems and a tool
supporting application of this context pattern. For the
risk assessment itself, we provide a different graphi-
cal modelling approach that focuses on the composi-
tion of the system in terms of assets and their interde-
pendencies. The tool in this case supports creation of
the system model, and used with a knowledge base of
security threats and controls, allows auto-generation
of a threat catalogue for the modelled system. The
tool then allows controls to be specified and risk lev-
els automatically calculated. The knowledge base in-
cludes security threats, compliance threats (models of
selected requirements under the GDPR), and also po-
tential modelling errors which are treated as threats
and highlight where aspects of the modelled system
may be underspecified or inconsistent.
In our methodology the risk model is usually de-
veloped first, capturing the expected use of software
and services from their suppliers. The context model
is captured later by the operator of the system, en-
coding information about the intended deployment,
which is then fed into the risk model which can then
be used to complete the risk assessment procedure.
The created models document the assumptions made
by stakeholders at each step and facilitate communi-
cation. These models, with a suitable formatting, can
provide documentation required for other purposes,
e.g. for ISO 27001 audits, or to generate notices or
policies on the use of personal data within the system.
The remainder of the paper is structured as fol-
lows: In Section 2, we briefly introduce the under-
lying concepts. Section 3 presents the contribution of
this paper by outlining the objectives, the models used
and our proposed methodology. Section 4 describes
the application of the methodology with an example.
We discuss related work in Section 5 and conclude our
paper with an outlook on future research directions in
Section 6.
2 BACKGROUND
In this section, we briefly introduce the fundamental
concepts for our risk assessment methodology.
Data Protection Goals. Data protection goals are
the following: confidentiality involves preventing
unauthorized access to data, integrity preventing
unauthorized or accidental alteration or corruption of
data, availability ensuring authorized access to data is
possible within an acceptable time, unlinkability sep-
arating privacy-relevant data from other data, trans-
parency ensuring all involve parties understand the
privacy-relevant data handling processes, and inter-
venability allowing involved parties to interfere with
processing to make corrections and prevent inappro-
priate processing (Zwingelberg and Hansen, 2011).
The General Data Protection Regulation (Euro-
pean Union, 2016) obliges anyone collecting or pro-
cessing data within the European Union to take re-
sponsibility for managing risks to Personally Identifi-
able Information (PII). Article 25 in particular stipu-
lates that the data controller shall “both at the time of
the determination of the means for processing and at
the time of the processing itself, implement appropri-
ate technical and organizational measures” to protect
PII, “taking into account the state of the art, the cost
of implementation and the nature, scope, context, and
purposes of processing as well as the risks of varying
likelihood and severity for rights and freedoms of nat-
ural persons posed by the processing”. In other words,
data protection goals should be achieved by assessing
risks during the design and operation of a system for
processing PII, and introducing measures to manage
those risks.
This poses considerable challenges for the reasons
discussed above. A data controller should assess and
manage risks to personal data, but they depend on
suppliers of software and services, and in cloud based
applications, even the management of resources used
to implement the data processing system. It is dif-
ficult for a data controller to identify potential threats
that should be managed, and ensure that the “technical
and organizational measures” in place are sufficient to
manage risks from those threats. This becomes more
challenging when one considers that it is not enough
to do this during the design of the system, but also
during its operation. In principle, one must reassess
risks whenever the system changes, or whenever new
types of threats are discovered.
ISO 27001. The normative ISO 27001
(ISO/IEC 27001, 2017) standard specifies “the
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
54
requirements for establishing, implementing, main-
taining, and continually improving an information
security management system within the context of an
organization”. Since these requirements are defined
in a generic way, the standard can be applied to
organizations of any business area. Annex A lists
information security control objectives and controls
that should be considered and used where relevant
in systems falling within the scope of the ISMS.
ISO 27001 also requires the use of a risk assessment
procedure to determine the appropriate level of risk
treatment using security controls from Annex A. The
scope of the ISMS should be defined by a Statement
of Applicability.
A typical response to GDPR Article 25 by data
controllers is to use ISO 27001 accredited suppliers
of software and operators of services and cloud re-
sources used in creating the personal data processing
system. In doing this, the data controller must de-
termine from the Statement of Applicability whether
the software or services they are using are covered by
an ISMS, and whether the risk management criteria
and assumptions used in those ISMS are sufficient for
the data controller’s purposes. Complex interdepen-
dencies exist between these externally sourced assets,
making it extremely difficult to do this in practice.
ISO 27005. The ISO 27005 (ISO/IEC 27005, 2018)
standard provides informative guidelines for con-
ducting an information security risk management in
an ISO 27001 compliant ISMS. Here, the different
guidelines give support for the application of an in-
formation security risk management in the different
phases.
An iteration of the information risk management
process starts with a context establishment, which in-
cludes definition of the scope and boundaries of the
system in which risks are assessed, and selecting:
i) the risk management approach and assessment if
the necessary resources for the risk management are
available, ii) risk evaluation criteria for the assessment
of information security risks (e.g. value of assets for
the business processes of an organization), iii) impact
criteria for assessing the impact (e.g. monetary loss)
if a security property of an asset gets compromised,
and iv) risk acceptance criteria, with respect to a scale
of risk levels.
After the context establishment, the risk assess-
ment starts with the risk identification, which involves
the following processes: 1) identification of the as-
sets within the defined scope, which may be difficult
if some assets are supplied or operated by third par-
ties, 2) identification of threats for the identified as-
sets, which is often easier for the suppliers or oper-
ators than for the data controller, 3) identification of
existing controls that are already implemented, which
is at least partly determined by the suppliers, espe-
cially for assets implemented by software, 4) identifi-
cation of vulnerabilities regarding the set of identified
assets, which depends on the suppliers or operators
of the assets, and 5) identification of the damages and
consequences to an organization if a security property
of an asset gets compromised.
The next step is risk analysis, which involves com-
prehending the nature and level of risk posed by the
identified threats. This is found by first estimating the
likelihood of each threat, given the presence of con-
trols and vulnerabilities, and the impact of the threat
given the harm caused should its consequences occur.
ISO 27005 defines two categories of assets: i) pri-
mary assets, i.e. information and business processes;
ii) supporting assets on which the primary assets rely,
including hardware, software, and non-technical el-
ements such as system administrators, etc. For pri-
mary assets, the impact is based on the direct conse-
quences to the organization (or in the case of PII, to
the data subject). For secondary assets, the impact of
any compromise depends also on the interdependen-
cies between assets. This is usually difficult to deter-
mine, as it is easy to overlook dependencies especially
when the assets are supplied or operated by third par-
ties.
The last step is risk evaluation, in which the accep-
tance criteria are applied to determine whether risk
levels are acceptable. If not, one must select a risk
treatment which may include avoidance (not using the
system or function in which the risk arises), trans-
fer (assigning responsibility to another party, although
this is not permitted under the GDPR), or risk reduc-
tion (introducing more security controls to reduce the
likelihood and/or impact of the associated threat).
3 RISK ASSESSMENT AND
EVALUATION
METHODOLOGY
In this section, we describe our methodology, which
considers risk assessments at different phases in the
lifecycle of a cloud-based software/service. Our
methodology provides an underlying conceptual
model, a knowledge base for the provision of a risk
assessment and guidelines for the different steps of a
risk assessment. It also includes tasks for the evalu-
ation of the implemented software/service and its de-
ployment by performing penetration tests.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
55
Service
Operator
Software/
Service
Supplier
Business and
Deployment
Context
Cloud
Resource
Provider
Software/Service
Security
Properties
Infrastructure/
Plattform Security
Properties
Service Operator
Risk Assessment
Figure 1: Organizational Dependencies in Risk Assess-
ment.
3.1 Objectives of Our Methodology
Today, cloud-based application software and services
are often designed and implemented by specialist sup-
pliers, independently from their eventual operator(s).
The former hold most of the knowledge required to
analyze information security risks, and are in a posi-
tion to introduce controls to address many of those
risks. Yet they lack information about the deploy-
ment environment and so cannot perform a complete
risk assessment, which ultimately must be done by the
service operator. This leads to dependencies between
different organizations in a risk assessment, as shown
in Figure 1. This shows the involved stakeholders: the
software/service supplier, the service operator who
manages the deployed system, and the provider of the
underlying cloud resources on which the system is de-
ployed. The service operator is responsible for the fi-
nal risk assessment, and needs information from the
other two stakeholders. This is not addressed explic-
itly by ISO 27005 as it describes the procedure to be
used within a single organization.
To solve these challenges, our methodology has
the following objectives: 1) provision of a concept
for a risk assessment that considers different lifecycle
phases of a software/service; 2) provision of guide-
lines for the execution of the different risk assessment
tasks; 3) provision of a knowledge base that provides
necessary information for the execution of risk assess-
ment tasks; and 4) provision of guidelines for the eval-
uation of the software/system by performing penetra-
tion tests.
Our approach depends on the creation and use of
models to define the system in which risks are be-
ing assessed and to support identification and anal-
ysis of risks within that system. Models are used be-
cause they provide three significant benefits compared
to other methods: 1) models provide an unambigu-
ous description of the system, the assumptions made
about potential sources of threats (e.g. attackers) and
the presence of security controls, etc., 2) models al-
low automated threat identification and risk analysis,
which is both repeatable and less likely to overlook
important risks, and also faster and better suited to
meet the GDPR obligation to frequently reassess risks
during the operation of a personal data processing
system, 3) the automated risk analysis can incorpo-
rate the analysis of attack paths and secondary effects,
making it much easier to specify the impact of threat
consequences, especially for threats to secondary as-
sets.
3.2 Underlying Models and Patterns
Context Modelling. The context of a system repre-
sents a specific part of the system environment. This
part is relevant for the definition of a system and un-
derstanding of its requirements (Pohl, 2010). A sys-
tem context includes stakeholders, technical compo-
nents, processes, events, and regulations (e.g. laws)
that are relevant for the system (Pohl, 2010).
A context model is a general structural description
of a particular type of system including its environ-
ment (cf. (Context-Patterns, 2018)). It enables the
description and representation of a concrete system
that represents an implementation of the system type
as documented in a model (with a graphical represen-
tation).
System Modelling. Modelling of cloud systems is
still in its infancy. Traditional modelling approaches
are insufficient as they focus mainly on the techni-
cal aspects of a system, and system functionalities.
Furthermore, they do not include information on the
deployment, and relation between different types of
technical parts of systems, e.g. hosts of applications.
These models are designed to help system designers
to graphically identify and analyze the threats that can
arise in a system (Broy et al., 2012; Lock and Som-
merville, 2010).
In our methodology, a system model represents
a system in which risks are being assessed in terms
of assets and dependencies. A system model is con-
structed from the asset classes defined in a knowl-
edge base, also called a domain model since it de-
scribes generic asset types in a given domain, such
as cloud services. To model cloud applications, the
domain model includes technical assets (physical or
virtual devices and software components such as web
applications, databases, etc), logical assets that rep-
resent capabilities or dependencies arising from as-
set interactions, and social or physical assets includ-
ing physical spaces and human stakeholders. The do-
main model also contains models of potential threats
to those assets, and security controls that could be
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
56
Context Establishment
Identification of assets
Identification of threats
Identification of existing
controls
Identification of
vulnerabilities
Identification of
consequences
Risk Analysis
Risk Evaluation
Risk Treatment
Risk Identification
Risk Assessment
ISO 27005 Standard
Artifacts of our method
Context Model
System Model
Figure 2: Our Approach and Relation to the Steps in ISO
27005 (ISO/IEC 27005, 2018).
used to protect the assets.
The design-time system model is normally created
by the software/service developers, describing the ex-
pected arrangement of services and their relationships
to an assumed cloud deployment environment. The
deployment model is created by the service operator,
based on the design-time model but characterizing the
actual cloud deployment environment.
3.3 Our Methodology
Our methodology is based on ISO 27005, and covers
the definition of the context, identification of assets,
threats and existing security controls, and the con-
sequences and impact of threats, and the subsequent
analysis of risks from these threats. The models de-
scribed above are used to support these processes as
shown in Figure 2.
The types of assets to be considered in cloud ap-
plications, divided into primary and secondary assets
as defined by ISO 27005, are shown in Figure 3.
This diagram shows only the asset types specified (as-
serted) explicitly by a user, which include primary
assets in the form of Information and Business Pro-
cesses, and a wide range of secondary assets includ-
ing Location, Network, Stakeholder, Hardware, and
Software. Location assets are used to denote physi-
cal locations, e.g., a company’s building, from which
we can infer risk levels from physical attacks or at-
tacks involving local access to devices. Network as-
sets represent different means by which devices can
Portable
Medium
Storage
Location
Asset Owner
Organization
Human
Processing
Peripherals
Local
Medium
Host
Network
Software
Hardware
Business Process
Information
Supporting Asset
Primary Asset
Stakeholder
Asset
responsible for
Figure 3: Conceptual Model for Asset Classes (Adapted
from (Gol Mohammadi et al., 2019)).
communicate, and allow inference of threats from net-
work attacks at OSI Layer 2 (which may depend on
the type of network) or Layer 3, as well as attacks
on specific types of asset interactions (e.g. attacks on
web communications). Devices provide the means to
store and process data, and allow modelling of threats
against both the devices and the supported processing.
Thus a Host asset can host a Process asset (represent-
ing one or more executing software components), or
store data assets representing information.
Stakeholders can be used to represent both human
and organizational stakeholders, and allow their rela-
tionships to asset to be captured. A human can man-
age devices or (implicitly) all devices at a location,
interactWith processes which means the human has
control over the process, and data assets can relateTo
a human, indicating the data is Personally Identifi-
able Information and the human is the data subject.
In the current version of the domain model, there are
also specialized subclasses of information represent-
ing the various Special Categories of data defined by
the GDPR Article 9, and subclasses of Human to rep-
resent adults and children, the latter being afforded
extra protection under the GDPR Article 8.
Other assets including logical assets representing
capabilities or dependencies within groups of assets
are automatically inserted by the tooling, based on
rules specified in the domain model itself.
The identification of threats, potential controls and
threat consequences are all handled automatically by
the tooling using knowledge encoded in the domain
model. However, these steps are supported explicitly,
and users still have to specify which of the potential
controls are already present in the system design or
the deployed system.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
57
(I) Design-time risk
assessment for
software/service
(II) Independent penetration
testing of software/services
during development
(III) Risk assessment at
deployment
(IV) Penetration testing
during deployment
(optional)
Software/service
supplier
performed by
Independent
penetration tester
performed by
Service operator
performed by
Service operator
performed by
Figure 4: Overall Approach of Our Methodology.
The domain model captures vulnerabilities as either
intrinsic (purely internal) or extrinsic (externally ex-
ploitable) flaws in asset character or implementation
(e.g. the potential presence of software vulnerabili-
ties), or the absence of controls (e.g. failure to au-
thenticate users of an asset). Because of this, identi-
fication of vulnerabilities is handled invisibly, and is
not supported explicitly by the tooling. The user only
has to define their level of trust that assets are free of
intrinsic or extrinsic flaws.
3.3.1 Overview of Our Methodology
The overall procedure for using the methodology cov-
ering both the design time and deployment risk as-
sessments is shown in Figure 4.
This shows our overall methodology with four
main tasks that are performed during different phases.
It starts with a risk assessment during the de-
sign/development of a cloud-based software/service
that is performed by personnel of the software/service
supplier ((I) in Figure 4). After the development of
the software/service, it is tested by independent pene-
tration testers to verify the results from the risk anal-
ysis ((II) in Figure 4). When the software/service
is supplied to a certain organization for its deploy-
ment, a further risk analysis is performed that consid-
ers characteristics specific to the deployment environ-
ment ((III) in Figure 4). This risk analysis is executed
by the service operator. If the service operator consid-
ers it as necessary and if it is feasible, further penetra-
tion tests should be performed on the software/service
within its actual deployment environment.
General Risk Analysis of Our Methodology. The
risk assessment steps (I) and (III) in Figure 4 are de-
(1) Create context model
(2a) Create system
model including assets
(2b) Refine the system
model with supporting
assets
(4) Identify threats
(9) Evaluate risk
Risk
acceptable?
no
yes
Context model
System model
including primary
assets
System model including
threats for assets
System model
including risk levels for
each threat
System model with
indicated unacceptable
risks
System model including
risk treatments
System model including
all assets
Artifacts of our method
yes
Assessment
satisfactory?
no
(3a) Identify impact for
primary assets
System model including
impact values for
primary assets
(3b) Identify impact for
supporting assets
System model including
impact values for
supporting assets
(6) Identify vulnerabilities
System model in which
threats that are not
mitigated are identified
(5) Identify
existing/planned controls
System model including
implemented/planned
controls for assets
(7) Extract security
objectives
List of security
objectives
(8) Analyse risk
(10) Treat unacceptable
risk
Risk Assessment
Figure 5: Overview General Risk Assessment.
scribed in more detail in Figure 5, which shows the
steps to be performed and the input and/or generated
output of each step. In this section we focus on the
general procedure, without considering if the risk as-
sessment is performed during the design time or de-
ployment. Specific variants used in each of the lifecy-
cle phases are described later in Section 3.3.2.
The steps are as follows:
(1) Create Context Model. Before the actual
risk assessment, the scope and boundaries for the risk
analysis have to be specified. The scope represents
the starting point for the identification of the assets
of an organization that are relevant for the risk as-
sessment. For establishing the context, the charac-
teristics regarding the business of an organization are
considered. These characteristics are represented by
the following information: 1) relevant legal, regula-
tory and contractual requirements; 2) relevant internal
stakeholders that are interacting with the provided ser-
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
58
vice(s); 3) relevant locations; 4) high-level assets that
represent business processes of an organization and
information that is related to the business processes;
5) interfaces (cf. (ISO/IEC 27005, 2018), Section 7.3,
p. 7f).
In our methodology the scope is specified by cre-
ating a context model (see Figure 5, step 1). To sup-
port this step, we provide a context pattern for cloud
computing scenarios (see Section 3.2). This pattern
can be instantiated for the considered cloud comput-
ing scenario. Our context pattern specifies the rele-
vant types of context information for cloud comput-
ing scenario. During its instantiation, the pattern el-
ements of appropriate types are instantiated. In this
connection, the main business processes are repre-
sented in the context pattern by instantiated cloud
computing services of the levels Infrastructure as a
Service (IaaS), Platform as a Service (PaaS) and/or
Software as a Service (SaaS). The context pattern
instance also includes the processed data. Further-
more, relevant legislations and regulations in the form
of appropriate indirect stakeholders, direct stakehold-
ers of the considered cloud computing service and
any relevant locations are identified within the pat-
tern. Strictly speaking, the data contained in a context
pattern instance is an asset that is normally identified
during the identification of assets. Our methodology
identifies the data already within the context estab-
lishment because the types of processed data are nec-
essary for deriving relevant regulations.
The output of this step is a context model in the
form of an instance of our context pattern.
(2a) Create System Model Including Primary
Assets. Based on the scope that is defined by the
instantiated context pattern in step 1, the assets that
are considered within the risk assessment (see Figure
5, steps 2a and 2b) are identified. Regarding assets,
the ISO 27005 distinguishes between primary assets
and supporting assets. Primary assets comprise the
business processes that are contained in the scope as
well as the processed information and data. Support-
ing assets are those elements on that the primary as-
sets rely (e.g. software, hardware, network). The cur-
rent step (see Figure 5, step 2a) relates to the identi-
fication of primary assets by creating a system model
(see Section 3.2). Accordingly, processes and the pro-
cessed data are modelled in this system model. The
modelled processes are less abstract than the busi-
ness processes in the context pattern instance from
step 1. The system model defines different types of
processes that represent running applications that may
process, store or exchange data. Therefore, a process
in the system model contains always a supporting as-
set in the form of a software. This software has to
be taken into account when a process is considered
during the risk assessment. Relations between differ-
ent processes are also modelled in a system model.
Regarding the data, the system model distinguishes
between data with normal protection needs and sensi-
tive data that requires high protection needs. The data
shall accord to the data in the context pattern instance.
Jurisdictions for the data processing are also modelled
in a system model. These jurisdictions shall corre-
spond to identified legislations in the context pattern
instance.
The output of this step is a system model that con-
tains primary assets and jurisdictions for the data pro-
cessing.
(2b) Refine the System Model with Support-
ing Assets. This step (see Figure 5, step 2b) iden-
tifies the supporting assets for the primary assets
from step 2a (see Figure 5, step 2a). The follow-
ing types of supporting assets specified in ISO 27005
are considered: 1) software; 2) hardware; 3) networks
and their components; 4) personnel; and 5) sites (cf.
(ISO/IEC 27005, 2018), Annex B.1.1, p. 28).
Our methodology extends theses types of sup-
porting assets by public spaces and private spaces.
These spaces represent physical locations where de-
vices may be located, and subnets (notably radio sub-
nets) may be accessed. There is a specialized private
space called a DataCentre which represents a data
center containing network assets and devices to avoid
having to include these explicitly in the system model.
Virtual hosts and virtual networks can be provisioned
at a data center, without worrying about the mapping
to physical hardware.
The identified supporting assets are included into
the system model that has been created during step
2a. Within the system model, hardware, networks,
and network components are represented by different
types of network assets. For a network asset asso-
ciations to primary assets in the form of hosted pro-
cesses and processed, stored, and/or exchanged data
are identified. Associations that represent connec-
tions to or interactions with other network assets are
also modelled, and are used to determine attack paths
and secondary effect cascades during the later iden-
tification of threats for the identified assets in step 4
(see Figure 5, step 4). Associations with human assets
(representing user roles) are used to model threats in-
volving malicious users (insider attacks), and threats
from user error, such as phishing attacks.
Relationships with relevant jurisdictions are also
taken into account, primarily to allow the detection of
cross-border data transfers and potential compliance
threats, e.g. where nationally regulated data (as de-
fined in the GDPR Article 9.4) moves across a border.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
59
The output of this step is the system model that
includes all primary and supporting assets.
(3a) Identify Impact for Primary Assets. The
impact measures the consequences if an asset gets
compromised by a threat. In this step (see Figure 5,
step 3a) the impact for primary assets is determined.
The knowledge base within our methodology contains
predefined default values for the impact level of each
type of compromise and asset type. These default
levels are mostly quite low except for sensitive data
classes, because in practice this is appropriate for sec-
ondary assets, and they can be of any type apart from
data.
(3b) Identify Impact for Supporting Assets.
This step (see Figure 5, step 3b) determines the im-
pact values for supporting assets. However, this step
is usually not required, because our model-based ap-
proach makes it possible to automatically detect and
analyze attack paths and secondary effect cascades.
Since the main business value of a secondary asset is
to support primary assets, it is often sufficient to as-
sume there is no direct impact from a compromise in
a secondary asset other than the propagation of con-
sequences to primary assets.
The output of step (3) is the system model includ-
ing the specification of impact for (mainly primary)
asset compromises.
(4) Identify Threats. In this step, possible threats
to the assets in the system model are identified (see
Figure 5 step 4). To this, a catalogue of threats that
refer to the different types of assets is provided. This
catalogue is generated using the threat identification
rules incorporated into the knowledge base. If new
types of attacks are discovered, which are not merely
the old attacks exploiting some new programming er-
ror, the knowledge base must be updated.
The output of this step is the system model includ-
ing a threat catalogue describing threats to each sys-
tem asset.
(5) Identify Existing/Planned Controls. Con-
trols are measures that have been or are planned to
be realized for assets to decrease the likelihood that
an asset gets compromised by certain threats. Thus,
the identification of existing/planned controls is a part
of the risk assessment (see Figure 5, step 5). Our
methodology supplies a catalogue that contains con-
trols for the different types of assets. This catalogue
is generated based on the knowledge base. During
the control identification, the existing and/or planned
controls can be selected from the set of controls that
are associated to a certain type of asset. Within our
methodology, controls that could address a particu-
lar threat are combined as so called control strate-
gies. When all the controls needed for a control strat-
egy are selected, the effect is to reduce the likelihood
of a threat causing the expected consequences. Each
control strategy has a strength parameter which deter-
mines the maximum residual threat likelihood when
the control strategy is selected. If new security mea-
sures are developed to protect certain types of assets,
these should be added to the knowledge base.
The output of this step is the system model in
which the appropriate implemented and/or planned
controls are specified at each system asset.
(6) Identify Vulnerabilities. As discussed above,
in our methodology this step is implicit, so no user ac-
tion is needed. A vulnerability associated with flaws
in assets is modelled in terms of asset trustworthiness,
while the lack of security measures is captured by the
presence (or in this case lack) of the relevant controls.
(7) Extract Security Objectives. This step con-
siders the extraction of security objectives (see Figure
5, step 7) for the considered software/service. It is not
adapted from the ISO 27005, but is included because
security objectives can be used as an input to any pro-
cesses used to verify security properties, including
penetration tests that may be applied in steps (II) and
(IV) in Figure 4. They can also be used to formulate
a designated level of security for a software/service to
third parties (e.g. cloud service providers). The se-
curity objectives are derived from the impact factors
specified in step (3), and the threats that potentially
cause high impact asset compromises as determined
from step (5).
The output of this step is a list of security objec-
tives.
(8) Analyze Risk. In the risk analysis (see Fig-
ure 5, step 8), for any identified threat regarding each
asset the risk level is assessed. A risk level is a quan-
titative value that represents the risk that an asset gets
compromised by the considered threat. It is assessed
based on values that have been identified during the
previous steps of the risk identification (see Figure 5).
Within our methodology, risk levels are determined
for any threat. A risk level is determined based on the
likelihood that a threat occurs and has the expected
consequences, in the presence of selected controls,
and the impact of these consequences taking account
of attack paths and secondary effects.
The output of this step is the system model in
which risk levels are assigned to the threats to each
asset.
(9) Evaluate Risk. During the risk evaluation (see
Figure 5, step 9), it is evaluated if the determined risk
levels from the risk analysis are acceptable referring
to the defined risk criteria. In our current domain
model, risk levels are expressed using a 5-point scale
from Very Low to Very High, and we suggest that on
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
60
this scale a Medium level risk might be acceptable for
a short time, but normally the worst case risk level
from any threat should be Low or Very Low. However,
this depends on the risk appetite of the user, who must
make a decision after step 9, as shown in Figure 5. If
any the level of any risk is too high, it must be treated
in a further step.
The output of the risk evaluation is the system
model in which unacceptable risk levels are indicated.
After the risk evaluation, it has to be evaluated if
the previous results of the risk assessment are mean-
ingful (see Figure 5). If this is not the case, the scope
of the risk assessment has to be adjusted and the steps
of the risk assessment have to be repeated. If the re-
sults are meaningful, the approach continues with the
treatment of unacceptable risks. Our domain model
and tooling provides some assistance with this, as po-
tential system modelling errors are also modelled as
threats, so if there are any modelling error threats in
the generated catalogue, it may mean assets or rela-
tionships have been overlooked, and the model may
need to be revised and the process from step (2) re-
peated.
(10) Treat Unacceptable Risk. The risk treat-
ment (see Figure 5, step 10) includes all risk levels,
which have been determined during the risk evalua-
tion, that are not acceptable. In our methodology (see
Figure 5 step 10), additional controls for the corre-
sponding asset can be selected with the objective to
reduce the risk level. At this, appropriate controls that
are provided by the knowledge base are selected or
new controls are added to the knowledge base and are
selected afterwards. After the selection of controls,
the steps for the risk analysis and risk evaluation for
the concerned assets are repeated to examine if risk
treatment has been resulted in acceptable risks. If this
is not the case for certain assets, these assets must un-
dergo a further risk treatment and so on.
It is also possible to use other risk treatments,
such as avoiding risk by not providing certain features
and corresponding assets by the software/service any
more. The concerned assets are removed from the
system model, or (if the removal is only temporary)
marked as disabled (this is included in the domain
model as a potential security control). Outsourcing
the risk is also possible.
The output of this step is an updated system model
with the selected risk treatments.
3.3.2 Variants
(I) Risk Analysis at Design-Time/Development.
At design time, the above procedure is used by the
developer of the software/service, ideally during its
design but certainly before offering it for use in a
specific application (see Figure 4 step I). The follow-
ing discussion just considers specific differences com-
pared to the general risk assessment (see Figure 5).
The context model created during the context
specification contains all context information that is
relevant for the software/service at the design time
and has to be considered during the development of
the software/service. Additionally, context informa-
tion can define requirements to a potential deploy-
ment environment that have to be fulfilled by a corre-
sponding cloud infrastructure (e.g. computing center
shall be located in the European Union because the
General Data Protection Regulation shall apply).
Within the identification of assets (see Figure 5
steps 2a and 2b), the primary assets represent the pro-
cesses of the software/service that host parts of the
software or applications of the service for process-
ing, storing, and/or transferring data. The affected
data itself is also considered as primary asset. The
supporting assets include components that are neces-
sary for the provision of the software/service. The
software/service developer may not know the precise
specification of some of these assets, but can encode
their assumptions by their choices at this stage. For
example, they may know that the service will be de-
ployed at a data center, and assume that processes will
run and data will be stored on virtual hosts. In other
cases, the model should be as general as possible, e.g.
if it is possible for one person to fulfil two roles, the
roles should not be combined because in general, two
different people may be involved. Other aspects may
be omitted altogether, and left for the service operator
to define, e.g. the jurisdictions in which services will
be deployed, and from where they may be accessed.
Identified controls (see Figure 5 step 5) for the
primary assets and supporting assets that are part of
the software/service shall be implemented during its
development. The controls for the other supporting
assets represent planned controls that shall be imple-
mented by a potential deployment environment. The
software/service developer can only implement con-
trols that are embedded in their components, but may
at this stage specify their assumptions or requirements
on which other controls shall be used.
The same applies to security objectives. Security
objectives for primary assets and secondary assets as
a part of the software/service shall be implemented
during the development. The security objectives for
the other supporting assets are planned and shall be
implemented by a potential deployment environment.
(II) Independent Penetration Testing of Soft-
ware/Services. This task (see Figure 4 step II) in-
cludes the specification and execution of penetration
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
61
tests regarding the software/service. These tests are
performed on exposed software/service endpoints and
interfaces to confirm that:
the strength of security controls is sufficient, e.g.
encryption algorithms and key lengths
potential vulnerabilities cannot be exploited, e.g.
using injection attacks, cross-site scripting at-
tacks, etc.
Test scenarios may be based on the risk model from
the design time risk analysis (see Figure 4 step I). For
example, if a software asset was specified as highly
trustworthy, one can perform a what if analysis by re-
ducing the trustworthiness level of this asset, and see-
ing which types of threats then lead to high risk lev-
els. The tests should then check for flaws that could
enable those types of threats to be carried out.
(III) Risk Assessment at Deployment. The risk as-
sessment in this task considers the deployment of the
software/service in an actual cloud environment (see
Figure 4 step III). This risk assessment is performed
based on the context model and system model from
the risk assessment at design time (see Figure 4 step
I). In the following, only specific differences com-
pared to the general risk assessment (see Figure 5) are
considered.
The context model from the context definition (see
Figure 5 step 1) is expanded by context information
that is specific to a certain deployment environment
(e.g. cloud provider, location of computing center,
service level agreements).
In the system model, further supporting assets (see
Figure 5 step 2b) that are specific to the deployment
environment are identified (e.g. components that are
used for virtualization) and have not been known at
design time. Additionally, supporting assets that have
been represented during design time in an abstract
way are concretized (e.g. the concrete type of a virtual
machine or router is specified). Because all primary
assets have been identified during the risk assessment
at design time, there is no need for a further identifi-
cation during deployment.
For the new added supporting assets impacts are
assessed (see Figure 5 step 3b), threats and counter-
acting existing controls are identified (see Figure 5
steps 4 and 5), vulnerabilities are identified (see Fig-
ure 5 step 6), security objects are extracted (see Figure
5 step 7), and a risk analysis (see Figure 5 step 8), a
risk evaluation (see Figure 5 step 9) and if necessary
a risk treatment are performed (5 step 10).
Regarding concretized supporting assets it has to
be checked if further threats have to be considered
(see Figure 5 steps 4). If this is the case, for these
threats existing mitigating controls have to be identi-
fied (see Figure 5 steps 5) and the remaining steps for
the risk assessment (see Figure 5 steps 6 - 9)) and if
necessary a risk treatment (see Figure 5 step 10) have
to be performed.
With regard to planned controls for supporting as-
sets, that have been identified during design time, it
has to be checked if these controls are actually imple-
mented by the deployment environment in a sufficient
way (see Figure 5 step 5).
(IV) Penetration Testing during Deployment. In
some cases, the service operator may wish to conduct
additional penetration tests to verify security in the
specific operational deployment environment. This
may be necessary where the software/service supplier
cannot arrange penetration testing in a sufficiently re-
alistic environment. This may arise if:
the service operator wishes to deploy soft-
ware/services in an environment that is unlike that
envisaged by the software/service supplier; or
the deployment environment is only accessible to
the service operator, e.g. because it uses a hybrid
cloud composed partly of in-house resources.
In these cases, if the service operator needs to ver-
ify security properties, they should carry out their own
tests, where permitted by the cloud (or other) resource
providers.
Any test scenarios are based on the results from
risk assessment at deployment (Figure 4 step III).
4 APPLICATION EXAMPLE
The methodology was tested with a commercial soft-
ware and service supplier, one of the partners in the
collaborative project in which it was developed. The
company involved is an SME, specializing in the pro-
vision of applications in health and social care mainly
for local government organizations. They provide
software in two ways: by deploying it in the cloud
and offering it as software as a service (SaaS), or by
supplying it to service operators who deploy it in their
preferred environment.
To avoid any potential ethical issues, the vali-
dation exercise reported here is based on a generic
data collection and analysis application, in which a
user collects data using an app running on their smart
phone, and sends it to be stored and analyzed by ser-
vices running in the cloud. Results of the analysis are
made available via a web portal. The SME often uses
this design, and depending on the application it may
be used to capture personal data related to the phone
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
62
Figure 6: Design Time Model of a Generic Data Collection and Analysis Application.
user, and to support self-analysis by that user or anal-
ysis by others, e.g. health and safety professionals.
The design time model created by our partner for
this application is shown in Figure 6.
The user of the phone is in the lower-left quad-
rant, using the app to collect data and submit it via
the API gateway process which is in the middle of the
diagram. The analyst is shown top-left, and both they
and the user have access to results from the data anal-
ysis via a browser running on their device. The anal-
ysis services are hosted by a data center, and use re-
sources modelled on those supported by Amazon Web
Services, although of course the cloud provider might
not be Amazon.
The figure shows the model after the risk assess-
ment procedure has been used to iteratively identify
and address unacceptable risks. Various controls have
been specified in the model, and at this stage risk
levels are all Low or Very Low. At this stage the
software/service provider may run penetration tests
to verify their components, or have an independent
penetration testing organization conduct such tests if
their customers need independent verification of com-
ponent security and trustworthiness.
A customer organization wishing to use the soft-
ware would take the design time model as a starting
point, including security control specifications and
trustworthiness assumptions. They would then ana-
lyze the intended deployment context, and modify the
risk model for their own purposes. The result of this
process is shown in Figure 7.
The differences between these models reflect the
specific deployment plans of the service operator. A
major change is that the data collected by the user is
now personal data relatedTo that user. The analyst
role has also gone, and the user now manages both
client devices and browsers, because in this case the
intention is to provide a self-analysis service to users.
Jurisdictions have been added showing that the opera-
tor is planning to deploy the service at a data center in
one country, and serve clients in a second country. Fi-
nally, the operator has a lower opinion of their cloud
service provider than the software/service developer
has of Amazon, which they modelled by reducing the
trustworthiness of the sysadmin role as shown.
These changes lead to two main differences in the
assessment of risks. Firstly, a number of GDPR com-
pliance threats will have been added to the threat cat-
alogue representing technical requirements from the
GDPR (in addition to the requirement to assess and
manage risks). These include the need for a legal
basis for processing personal data, which the oper-
ator decided to do by consent from the user. This
means the process of registering to use the app must
include a suitable notice allowing the user to express
(or withhold) their consent. Secondly, because the
cloud sysadmin role is considered only moderately
trustworthy, risk levels from insider attacks are much
higher, leading the operator to specify that trusted
hardware must be accessible by the provisioned VMs
so sensitive processes can be protected even from the
system manager. If this is not available, they may de-
cide to use their own in-house facilities instead of run-
ning this application in the cloud.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
63
Figure 7: Deployment Model for the Data Collection and Analysis Application.
5 RELATED WORK
There is a recent case study on threat modeling that re-
vealed the effectiveness and efficiency of threat mod-
eling in large scale applications (Stevens et al., 2018).
The participants reported that threat modeling sup-
ports the communication in teams and the develop-
ment of mitigation strategies. Since we provide a tool-
support methodology and graphical representations of
the models, we improve the scalability for larger ap-
plications.
The ISO 27005 standard considers three major
steps for risk identification and treatment: (i) identi-
fying assets, (ii) identifying threats and evaluate their
risks for the assets, and (iii) identifying controls to
protect the assets against threats. Existing risk man-
agement methods mostly do not consider explicitly
documenting and modeling the assets of the system
under consideration.
Lund et al. propose the CORAS risk management
process (Lund et al., 2011). The authors describe the
process of identifying threats as an interactive brain-
storming session. By considering our methodology,
the threat and control identification is systematic and
even can be used for structuring the brainstorming
sessions, if necessary. Furthermore, before threat and
control identification, using our methodology the con-
text and boundaries for performing risk assessment is
specified. We also provide more systematic way, for
identifying asset in a model name design-time system
model that include primary and supporting assets as
described in ISO standard.
The ProCOR method (Wirtz et al., 2018) describes
a risk management process which combines problem
frames with CORAS. In this method, the identifica-
tion of threats relies on external knowledge and lacks
of a systemization. The same authors propose an-
other method for identification of assets (Gol Moham-
madi et al., 2019). The approach considers a problem
frame model for the identification of relevant assets.
In contrast to their work, we propose a combined ap-
proach with context modelling and a risk assessment
method which enhance the asset identification. Fur-
thermore, the threats and control identification are not
only based on the expertise of security engineers, but
also tool support with a comprehensive knowledge
base of possible threats has been considered.
Abuse frames describe problem-oriented patterns
to analyze security requirements from an attacker’s
point of view (Lin et al., 2003). An anti-requirement
is fulfilled when a threat initiated by an attacker is re-
alized. Domains are considered as assets. The mali-
cious machine of an abuse frame acts as the interface
between the attacker and the asset domain. Compa-
rable to problem frames, abuse frames are patterns
to describe the typical behavior of attackers. To ap-
ply an abuse frame, it is composed of a base problem
which is represented by a problem frame. Composing
means to map domains from the base problem into the
abuse frame. Based on the composed abuse frame,
the attacker’s behavior can then be further analyzed.
Abuse frames consider problems in isolation, whereas
we consider the system as a whole together with its
context.
Haley et al. provides a problem-oriented frame-
work to analyze security requirements (Haley et al.,
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
64
2008). Using so called satisfactory arguments, the au-
thors analyzes a system with regard to security based
on its functional requirements. There is no specific
way to identify threats. Instead, our methodology al-
lows to analyze the system as a whole together with
its context, i.e. to analyze context domains, assets,
threats and controls in combination.
In the context of goal-oriented software-
engineering, Secure Tropos allows to to model
and analyze security requirements along with func-
tional requirements (Mouratidis and Giorgini, 2007).
Another goal-oriented approach is the usage of
anti-models (van Lamsweerde, 2004). Currently, we
do not consider goals in detail and do not provide
any goal refinement process. Instead, we put a
special focus on data protection requirements and the
corresponding regulations in context model and data
protection requirement in possible data flows which
we analyze with regard to information security.
Opdahl and Sindre introduce misuse cases as an
extension of use cases. A misuse case describes how a
software shall not be used, i.e. how potential attackers
may harm the software (Sindre and Opdahl, 2005).
Misuse cases are a high-level description which does
not provide any details about protection of assets, as
we do in our approach.
Microsoft developed a method called STRIDE
(Shostack, 2014). It is a popular security framework
which is used to identify security threats. Using data
flow diagrams for modeling the system and its behav-
ior, threats are elicited based on existing threat cate-
gories: Spoofing, Tampering, Repudiation, Informa-
tion Disclosure, Denial of Service, and Elevation of
privilege. Each of these categories is a negative coun-
terpart to a security goal.
The STORE method allows to elicit security re-
quirements in the earliest phases of a software devel-
opment process (Ansari et al., 2019). The authors fol-
low a similar approach to first identify domain knowl-
edge such as external entities and vulnerabilities. To
identify threats, the authors propose a threat dictio-
nary which contains a collection of previously iden-
tified threats. However, their work has no systematic
context analysis and asset identification prior to threat
identification.
The other innovation of our approach is that we
further take that gained knowledge from threat and
control identification to further enrich the context in-
formation of a cloud system. The revised version of
context model is beneficial during the deployment-
time. Furthermore, our approach is conform to ISO
standard. We employ context and asset models in our
methodology, i.e., by developing a tool-assist method-
ology that makes it possible to set up an context model
and asset model (named in this paper, design-time
system model). Furthermore, our methodology sup-
ports the automated identification of threats and con-
trols based on these models.
6 CONCLUSION
This paper described a methodology for applying an
ISO 27005 risk assessment procedure to cloud based
applications in which the operator of a service ob-
tains software/services from a third party developer,
and deploys using resources from a separate cloud re-
source provider. The methodology is based on the use
of models, and provides several benefits over conven-
tional approaches:
the models provide a means for the soft-
ware/service developer to contribute to the risk
assessment, and communicate their requirements
and assumptions unambiguously and in a useful
form;
the models allow context to be captured and in-
corporated by the service operator, enabling an as-
sessment of risks in specific circumstances;
the models ensure consistency and traceability, al-
lowing the relationship between security proper-
ties and controls and consequent risk levels to be
found and used when formulating security verifi-
cation tests;
it is possible to automate the identification of
threats and assessment of risk levels, making it
less likely that any significant threats (including
multi-step attack paths and secondary effects) will
be overlooked;
automation also reduces the time and cost to re-
peat a risk assessment, making it easy to address
different applications based on the same software,
or to analyze a system if there is a change in the
system or in the threat landscape.
In future, we aim to exploit our models even fur-
ther, moving beyond the deployment and facilitating
run-time risk assessments, so providing a means to
comprehensively address the requirement under the
GDPR Article 25 for security by design and by de-
fault, not only during the design of a system for han-
dling PII, but also “at the time of the processing it-
self”.
Systematic Risk Assessment of Cloud Computing Systems using a Combined Model-based Approach
65
ACKNOWLEDGEMENTS
This work received funding from the EU’s Horizon
2020 research and innovation programme under grant
agreement 731678 (RestAssured). We thank Roman
Wirtz for his help in preparing the final version of this
paper.
REFERENCES
Ansari, M. T. J., Pandey, D., and Alenezi, M. (2019).
STORE: security threat oriented requirements engi-
neering methodology. CoRR, abs/1901.01500.
Broy, M., Cengarle, M. V., and Geisberger, E. (2012).
Cyber-Physical Systems: Imminent Challenges, pages
1–28.
Context-Patterns (2018). Overview, http://context-
patterns.info/index.html.
European Union (2016). General Data Protection Regula-
tion. https://gdpr-info.eu.
Gol Mohammadi, N., Wirtz, R., and Heisel, M. (2019). Sys-
tematic asset identification and modeling during re-
quirements engineering. In 14th Intl. Conf. on Risks
and Security of Internet and Systems (CRiSIS).
Haley, C. B., Laney, R. C., Moffett, J. D., and Nuseibeh, B.
(2008). Security requirements engineering: A frame-
work for representation and analysis. IEEE Trans.
Software Eng., 34(1):133–153.
ISO/IEC 27001 (2017). Information technology - Security
techniques - Information security management sys-
tems - Requirements.
ISO/IEC 27005 (2018). Information technology - Security
techniques - Information security risk management.
Lin, L., Nuseibeh, B., Ince, D. C., Jackson, M., and Moffett,
J. D. (2003). Analysing security threats and vulnera-
bilities using abuse frames.
Lock, R. and Sommerville, I. (2010). Modelling and Analy-
sis of Socio-Technical System of Systems. In Proc. of
the 15th IEEE Intl. Conf. on Engineering of Complex
Computer Systems, ICECCS, pages 224–232.
Lund, M. S., Solhaug, B., and Stølen, K. (2011).
Model-Driven Risk Analysis - The CORAS Approach.
Springer.
Mouratidis, H. and Giorgini, P. (2007). Secure tropos: a
security-oriented extension of the tropos methodol-
ogy. intl. Jour. of Software Eng. and Knowledge Eng.,
17(2):285–309.
Pohl, K. (2010). Requirements engineering: Fundamentals,
principles, and techniques. Springer.
Shostack, A. (2014). Threat Modeling: Designing for Se-
curity. Wiley.
Sindre, G. and Opdahl, A. L. (2005). Eliciting security re-
quirements with misuse cases. Requir. Eng., 10(1).
Stevens, R., Votipka, D., Redmiles, E. M., Ahern, C.,
Sweeney, P., and Mazurek, M. L. (2018). The bat-
tle for new york: A case study of applied digital threat
modeling at the enterprise level. In 27th USENIX Se-
curity Symposium, USENIX Security, pages 621–637.
van Lamsweerde, A. (2004). Elaborating security require-
ments by construction of intentional anti-models. In
26th Intl. Conf. on Soft. Eng. (ICSE),, pages 148–157.
Wirtz, R., Heisel, M., Borchert, A., Meis, R., Omerovic, A.,
and Stølen, K. (2018). Risk-based elicitation of se-
curity requirements according to the ISO 27005 stan-
dard. In Evaluation of Novel Approaches to Software
Engineering - 13th Intl. Conf., ENASE, pages 71–97.
Zwingelberg, H. and Hansen, M. (2011). Privacy protection
goals and their implications for eid systems. In IFIP
PrimeLife Intl. Summer School on Privacy and Iden-
tity Management for Life, pages 245–260. Springer.
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
66