A Framework for Security and Risk Analysis of Enrollment Procedures:
Application to Fully-remote Solutions based on eDocuments
Marco Pernpruner
1,2 a
, Giada Sciarretta
1 b
and Silvio Ranise
1,3 c
1
Security & Trust Research Unit, Fondazione Bruno Kessler, Trento, Italy
2
Department of Informatics, Bioengineering, Robotics and System Engineering, University of Genoa, Genoa, Italy
3
Department of Mathematics, University of Trento, Trento, Italy
Keywords:
eDocuments, Enrollment, Onboarding, Risk Analysis, Security Analysis, Security Framework.
Abstract:
More and more online services are characterised by the need for strongly verifying the real-world identity of
end users, especially when sensitive operations have to be carried out: just imagine a fully-remote signature
of a contract, and what could happen whether someone managed to perform it by using another person’s
name. For this reason, the identity management lifecycle contains specific procedures – called enrollment or
onboarding – providing a certain level of assurance on digital users’ real identities. These procedures must be
as secure as possible to prevent frauds and identity thefts. In this paper, we present a framework composed
of a specification language, a security analysis methodology and a risk analysis methodology for enrollment
solutions. For concreteness, we apply our framework to a real use case (i.e., fully-remote solutions relying on
electronic documents as identity evidence) in the context of a collaboration with an Italian FinTech startup.
Beyond validating the framework, we analyse and highlight the essential role of mitigations on the overall
security of enrollment procedures.
1 INTRODUCTION
Verifying a person’s real identity has always been a
risky operation, considering that paper-based iden-
tity documents could be counterfeited with little or
even no possibility of detection. This is one of the
main reasons why electronic documents (also called
eDocuments) have been created: these documents
are equipped with contactless chips that can be read
wirelessly, and in some cases they are also capable
of performing cryptographic operations. In addition,
eDocuments are tamper-resistant and provide a high
level of assurance on the attested data. As a result,
many companies that previously required customers
to physically visit a branch to perform some sensi-
tive operations (such as opening a new bank account
or signing a contract) can now leverage the intrin-
sic security assured by eDocuments to provide fully-
remote procedures and thus improve the usability on
their clients’ side. Although NIST defines these pro-
cedures as enrollment (Grassi and Fenton, 2017), they
a
https://orcid.org/0000-0001-8936-2726
b
https://orcid.org/0000-0001-7567-4526
c
https://orcid.org/0000-0001-7269-9285
are commonly referred to as onboarding, especially
when they lead to customer acquisitions.
However, simply relying on eDocuments’ security
is not enough: enrollment procedures must be sturdily
designed and adopt suitable security mechanisms to
prevent impersonation attacks. Moreover, such pro-
cedures as well as physical ones are subject to
many guidelines and regulations that have been inter-
nationally issued to prevent frauds and identity thefts
(e.g., Anti Money-Laundering in the financial sector).
In this paper, we present a framework for secu-
rity and risk analysis of enrollment procedures, that is
based on the following components:
Specification Language: starting from the anal-
ysis of on-the-market solutions, we have derived
a list of basic actions that are commonly involved
in enrollment procedures. In order to allow an
easy description and analysis of the procedures,
we describe them by means of a simple specifica-
tion language with precise semantics;
Security Analysis Methodology: we describe a
security analysis methodology for enrollment pro-
cedures by defining the concept of identification
factors (that we have borrowed and adapted from
222
Pernpruner, M., Sciarretta, G. and Ranise, S.
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments.
DOI: 10.5220/0010554502220233
In Proceedings of the 18th International Conference on Security and Cryptography (SECRYPT 2021), pages 222-233
ISBN: 978-989-758-524-1
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
the authentication context), threat model and se-
curity goal;
Risk Analysis Methodology: we describe a risk
analysis methodology to provide a prioritization
of the attackers detected during the security anal-
ysis according to their severity.
Our framework can be generally adopted to de-
scribe and analyse any enrollment procedure, by sim-
ply adjusting if needed the set of actions in-
volved (i.e., the specification language) and the capa-
bilities of the attackers on these actions (i.e., the threat
model). For concreteness, we have applied the frame-
work in the context of a realistic use case (i.e., fully-
remote solutions relying on eDocuments), where we
have also analysed the role of security mitigations on
the overall security of enrollment procedures.
Structure of the Paper Section 2 provides some
essential concepts to understand what follows; Sec-
tion 3 identifies our requirements; Section 4 defines
our framework and its components; Section 5 dis-
cusses some security mitigations that we have iden-
tified; Section 6 shows the application of the frame-
work to a real use case and highlights the essential
role of security mitigations; Section 7 discusses some
state-of-the-art solutions considered throughout this
work and Section 8 finally draws conclusions.
2 BACKGROUND
Enrollment. The National Institute of Standards
and Technology (NIST) has recently released the Spe-
cial Publication 800-63 suite (NIST, 2017) dealing
with digital identity. According to the definitions in
(Grassi et al., 2017b; Grassi and Fenton, 2017), the
enrollment can be considered as the procedure allow-
ing users to obtain a credential from a specific ser-
vice, after the verification of their identity. An enroll-
ment flow is composed of three main parts: (i) Res-
olution: the service collects the personal information
of the applicant and some identity evidence (e.g., a
digital copy of an identity document through a cam-
era or a scanner); (ii) Validation: the service veri-
fies the genuineness of the identity evidence provided
(e.g., by querying authoritative sources or visually
checking the documents), thus obtaining guarantees
on the identity information; (iii) Verification: the ser-
vice checks that the identity evidence belongs to the
applicant (e.g., by collecting a picture of her and com-
paring it with the one embedded in the document). All
these parts are extremely sensitive and potentially ex-
posed to many security issues: in case Validation and
Verification were not properly secured, users could
provide wrong or fictional personal information, al-
tered identity evidence or even identity evidence be-
longing to someone else. Therefore, enrollment pro-
cedures must be properly designed to rely on cutting-
edge techniques and assure that the user is really who
she claims.
Electronic Documents. When dealing with paper-
based identity documents, the detection of tampered
or misused identity evidence is not trivial, since crimi-
nals have always found new ways to counterfeit them.
For these reasons, electronic versions of identity doc-
uments have recently been produced: passports be-
came electronic passports (ePassports), while elec-
tronic identity cards (eID cards) replaced paper-based
identity cards. These eDocuments are equipped with
a contactless chip that can be read wirelessly and con-
tains the citizen’s personal data; in this way, criminals
would need to forge not only the printed data, but also
those contained within the chip (that is much more
difficult, as they are digitally signed).
As for ePassports, to allow only authorized people
to access the personal data, the identity page contains
a machine-readable zone (MRZ) that must be scanned
by the reader to derive the key needed to mutually au-
thenticate with the ePassport’s chip; the reader can
then interact with the ePassport in order to get the
personal data. These data can be validated by check-
ing the Document Security Object (SOD), a specific
file containing the signature of all the personal data’s
hashes: the reader can thus verify the authenticity and
integrity of the data. This procedure, currently used
by customs and immigration offices at borders, is pro-
vided by the MRTD application (ICAO, 2015).
The MRTD application is also contained in eID
cards, which in addition provide another mod-
ule for online authentication and, indirectly, for iden-
tification: the IAS ECC application (GIXEL, 2009).
In fact, given the key pair and the X509 certifi-
cate (Cooper et al., 2008) they are equipped with,
eID cards can be used in a challenge-response pro-
tocol: an application willing to authenticate the user
can send a challenge (i.e., a piece of information asso-
ciated with the ongoing operation) to the reader, and
thus to the eID card. By signing the challenge through
its private key, the eID card generates the response;
this operation can be performed only once the user
has inserted the PIN of the specific document. Since
the eID card’s public key is known, the application
can verify the proper signature of the original chal-
lenge, finally authenticating the user.
.
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
223
Table 1: OWASP risk computation.
Likelihood
Low Medium High
| Low Note Low Medium
Impact Medium Low Medium High
High Medium High Critical
OWASP Risk Rating Methodology. The Open
Web Application Security Project (OWASP) has
developed the Risk Rating Methodology (OWASP,
2018), an approach to estimate the severity of spe-
cific risks. The risk is defined as likelihood × impact,
where the likelihood rates the probability of exploit-
ing a specific vulnerability, while the impact measures
the consequences of an attacker successfully leverag-
ing a vulnerability.
Computing the likelihood and the impact requires
the identification of some suitable factors, which can
be dependent on the considered scenario and have to
be given a score ranging from 0 to 9; lower scores for
likelihood factors mean less likely, while when deal-
ing with the impact they denote less serious conse-
quences in case the considered vulnerability was ex-
ploited. The official OWASP documentation provides
some examples regarding factors, although they can
be fully customised: for instance, Ease of Discovery
may be a likelihood factor representing how easy the
considered vulnerability can be discovered (where 1
= practically impossible, 3 = difficult, 7 = easy, 9 =
automated tools are available for this purpose).
Once these factors have been identified and rated,
the overall likelihood and impact result from the aver-
age of such scores, which are then assigned a qualita-
tive label according to their value: Low if value < 3,
Medium if 3 value < 6, High if value 6.
Finally, the overall risk can be found by combin-
ing the likelihood and impact labels as in Table 1.
3 REQUIREMENTS
Enrollment procedures are now playing a key role in
many sensitive contexts. In addition, the current situ-
ation connected with COVID-19 preventing people
from physically visiting public offices and company
branches further requires a way to be securely and
easily identified from home, possibly by using already
owned devices. One of the easier and less complex
remote identification methods consists of a video call
between a service operator and the person who has to
be identified, with the former asking the latter for any
document needed to finalise the operation. However,
since human operators have to be physically involved
in the call, this method may slow down the enroll-
ment process, especially when a considerable amount
of people needs to be identified in the same period.
Moreover, as online services should be accessible by
everyone, the enrollment procedure should be carried
out as simply as possible (e.g., without requiring peo-
ple to perform complex operations or to install soft-
ware). From all these considerations, we can infer the
following requirements that an enrollment procedure
should meet. The procedure must:
R1. be carried out remotely and automatically,
without human operators for identification;
R2. rely on devices that people already own;
R3. provide an adequate level of usability, thus
allowing everyone to finalise it.
4 FRAMEWORK
In this section, we describe the framework that we
have developed, composed of a specification language
to model enrollment protocols (Section 4.1), a secu-
rity analysis procedure (Section 4.2) and a risk anal-
ysis procedure (Section 4.3). Finally, we discuss the
usefulness of the framework (Section 4.4).
4.1 Specification Language
The specification language aims at providing a sim-
ple and graphical way to model enrollment proce-
dures. In this work, for the sake of concreteness, we
have decided to instantiate the requirements defined
in Section 3 by focusing on eDocuments, which allow
secure remote identifications (satisfying R1). More-
over, the automatic data extraction from eDocuments
enhances the overall level of usability and prevents
mistakes (satisfying R3). However, to benefit from
all these capabilities, people would need to own NFC
readers enabling the interaction with the eDocuments.
Instead of forcing people to buy a reader to plug into
their personal computers, we require a mobile appli-
cation (hereafter app) that can be used with smart-
phones or tablets equipped with NFC (satisfying R2).
Consequently, by focusing on eDocuments, we
have identified a list of basic actions that are com-
monly involved in enrollment procedures based on
such means, thus defining the specification language
displayed in Table 2. When modelling an enrollment
procedure, the involved actions are depicted in se-
quence separated by a semicolon; this is the visual
representation of an enrollment flow, which can be de-
fined as a finite set of actions from those in Table 2.
SECRYPT 2021 - 18th International Conference on Security and Cryptography
224
Some actions may require a basic entity as argument,
which is represented between parenthesis.
Example 1. An enrollment procedure requiring
users to insert the eID card’s PIN (÷(PIN)), read the
document through NFC ( (æ)) and finally confirm the
correctness of the extracted data (¡) can be mod-
elled as follows:
÷(PIN); (æ); ¡
Although for concreteness we focus on enrollment
procedures based on eDocuments, the specification
language can be used with any enrollment procedure
by properly extending if needed the set of involved
basic actions.
4.2 Security Analysis
In order to perform a security analysis, we refer to
the idea of factors. While authentication factors have
been defined by NIST (Grassi et al., 2017a) for au-
thentication, we are not aware of a similar notion for
enrollment. To fill this gap, in this work we intro-
duce identification factors. These factors are associ-
ated with the actions that the user is required to per-
form, though not all actions are necessarily associated
with an identification factor. For instance, ¯ is not
related to any identification factor, while ÷(PIN) is as-
sociated with
{
PIN
}
.
We define as security goal (SG) the set of identifi-
cation factors that should not be compromised for the
enrollment procedure to be considered secure.
Example 2. Considering the flow in Example 1, the
identification factors involved are the eID card and
the related PIN. Therefore:
SG =
{
æ, PIN
}
A threat model (T M ) over the identification fac-
tors is a pair (AT T , C ), where AT T is the set of con-
sidered attackers (detailed below) and C represents
their capabilities. More precisely, C is a collection
of mappings corresponding to each attacker in AT T :
given an attacker and an identification factor, C re-
turns a padlock representing the attackers’ effects on
that identification factor: the green padlock () de-
notes a non-compromised factor, while compromised
factors are represented by a red padlock (). More-
over, we use an asterisk (
) to depict a factor that is
compromised indirectly (e.g., in case an attacker de-
ceives the victim into interacting with her eDocument
instead of physically stealing it).
Example 3. Following Example 2:
a person standing behind the user could look at
the PIN while the user is typing (PIN ), but
he has no effect on the eID card (æ );
a malicious application on the user’s mobile de-
vice could directly intercept the PIN while the user
is typing (PIN ) and deceive the user into
placing the eID card near the device, thus indi-
rectly compromising it (æ
).
To avoid missing some attacks, C should be de-
fined through a worst-case approach, unless some se-
curity mitigations are implemented in the protocol; in
this case, their effects must be taken into considera-
tion when defining the capabilities in C. For instance,
a malicious application running on the user’s mobile
device should be considered able to escalate the de-
vice’s privileges and obtain root access, unless some
mitigations prevent this behavior or warn the user.
To identify the set of attackers for our analysis,
we take inspiration from the Enrollment and Identity
Proofing Threats defined by NIST (Grassi and Fen-
ton, 2017): among these, we do not consider attackers
belonging to the falsified identity proofing evidence
category, as they cannot be effective due to eDocu-
ments’ tamper-resistance, nor enrollment repudiation
that for the moment is out of the scope of this work (it
deals with an already enrolled user that denies having
performed an enrollment). Instead, we focus on the
fraudulent use of another’s identity category, whose
attackers aim to use identity documents belonging to a
different individual to complete an enrollment. There-
fore, in our analyses, AT T is the set of the following
attackers:
Identity Document Thief (IDT): steals an iden-
tity document from its legitimate owner;
Eavesdropping Software (ES): intercepts the
data typed on the device (e.g., keylogger);
Shoulder Surfer (SS): obtains secrets by looking
at the user inserting sensitive information;
Social Engineer (SE): exploits human gullibil-
ity and confidence in others, thus deceiving peo-
ple into revealing secret information or perform-
ing actions to their advantage;
Malicious Application (MA): runs on the at-
tacker’s or the victim’s mobile device. In the for-
mer case, the app can fake or alter the communi-
cation between the legitimate app and the server;
in the latter case, instead, the app can both inter-
act with the victim to obtain her personal data or
pictures, and access the device’s internal storage
to get the needed information autonomously.
An enrollment flow violates the security goal S G
under the threat model T M = (AT T , C) iff there is
an attacker (or a combination of them) in AT T that –
given its capabilities defined in C – compromises (di-
rectly or indirectly) all the identification factors con-
tained in the SG associated to the flow.
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
225
Table 2: Specification language for the description of the flows.
Basic entities
æ The eID card m The ePassport
ç An additional personal document PIN The PIN of the eID card
; The MRZ printed on the eDocuments The selfie captured by a user
Actions
The user may be required to...
¯ agree with the app’s privacy policy and terms
choose the type of eDocument she wishes to use
and the interaction mode
¥
provide some extra information that is not included
in the eDocument
¡
check and confirm the correctness of her personal
data, as extracted from her eDocument
@
insert her email address and to verify it
(e.g., through a link or an OTP sent by email)
insert her phone number and to verify it
(e.g., through an OTP sent by SMS)
capture a photo selfie. In case the selfie also needs
to contain an additional element, this will be speci-
fied as argument
()
place the element specified as argument near the
device, so that the app can interact with it through
NFC
¿
capture a video selfie ê()
scan the element specified as argument through the
device’s camera
() take a picture of the argument ÷() insert the information specified as argument
Example 4. Following Example 3, the malicious
application (now defined as MA) violates the security
goal SG since it manages to compromise all the iden-
tification factors involved (i.e., all these identification
factors are marked with a red padlock: or
).
Definition 1. The security analysis problem for an
enrollment flow under a threat model T M is to find
all (if any) minimal subsets AT T AT T so that ATT
violates SG.
Specifically, a subset AT T AT T is minimal iff
AT T violates S G and, for each ATT
0
( AT T , ATT
0
does not violate SG.
Example 5. Given Example 4, the considered
flow can be violated by
{
MA
}
; moreover, any fur-
ther combination involving MA (e.g.,
{
MA, IDT
}
or
{
MA, ES
}
) is obviously successful, since MA’s capa-
bilities are unchanged or even enriched. However,
only
{
MA
}
will be considered since the other attack-
ers do not represent minimal subsets.
4.3 Risk Analysis
The risk analysis that we perform is based on the
OWASP Risk Rating Methodology described in Sec-
tion 2. By taking inspiration from (Pernpruner et al.,
2020), in our analysis we consider the following like-
lihood factors
1
:
1
The OWASP Risk Rating Methodology uses the term
factors to identify the elements that are considered to com-
pute the likelihood and impact (cf. Section 2). It is highly
important to be aware that such factors have a completely
different meaning than identification factors.
Technical Difficulty (TD): rates the technical
difficulty of an attack to be successfully per-
formed. More difficult attacks are connected with
a lower likelihood;
Opportunity (O): provides a measure of the at-
tacker’s opportunity to perform the attack. This
opportunity can be restricted by limitations affect-
ing the chance of successfully completing an at-
tack; for instance, a device protected with a screen
lock would be more difficult (and thus less likely)
to be used than unprotected ones;
Attack Vector (AV): considers the means through
which the attack must be performed. An attack
requiring a physical intervention would be less
likely than a fully remote one;
User Interaction needed (UI): measures
whether an attack requires an interaction with
the victim to be successful. Attacks requiring a
precise operation by the user would be less likely
than others not requiring any interaction;
Spread of Attack (SA): deals with the spread of
the considered attack, according to trusted statis-
tics. Popular attacks increase the probability.
Instead, the impact factors are the following:
Attack Scale (AS): considers how many people
would be affected by the attack. More involved
users result in a higher impact;
Attack Detection (AD): deals with the possibil-
ity of detecting a successful attack. When an at-
tack is discovered, it can be hindered, thus making
easy-to-detect attacks have a lower impact.
SECRYPT 2021 - 18th International Conference on Security and Cryptography
226
In case some security mitigations are implemented
in the protocol, their effects must be taken into con-
sideration when assigning values to the factors above.
Definition 2. The risk analysis problem for an en-
rollment flow under a threat model T M = (AT T , C )
is to find the risk associated with all the minimal sub-
sets of attackers violating SG.
Example 6. The risk of a single attacker (e.g., MA
from Example 5) is determined as follows. First, we
need to assign a justified value to each likelihood and
impact factor; for instance, considering that MA can
operate remotely, its Attack Vector is 7. By repeating
this operation for each factor, we obtain all the val-
ues, which for MA can be the following: for the like-
lihood TD = 3, O = 1, AV = 7, UI = 1, SA = 2; for
the impact AS = 8, AD = 7. Then, after computing
the overall likelihood (2.80) and impact (7.50) by the
average of the corresponding factors, we can assign
them the corresponding labels (see Section 2): Low
to the likelihood, High to the impact. The combina-
tion of these values, according to Table 1, results in a
Medium risk.
When dealing with a combination of attackers,
finding a unique value for each factor is not trivial: in
fact, every attacker that is part of the combination has
his own values for all the factors. Therefore, we need
to combine attackers’ values as if we were consider-
ing a single attacker. Given the factors’ meanings, we
use the following logic:
for TD, AV, UI, AS and AD the combined value
should be the minimum of all the attackers’ values
for the considered factor;
for O and SA the aggregated value should be
less than the minimum of all the attackers’ values
for the considered factor.
Example 7. Considering the combination
{
IDT , ES
}
, the thief (IDT) requires physical in-
tervention (AV = 1) while the keylogger (ES) can act
remotely (AV = 7); therefore, even the combination
of such attacks must be performed (at least partially)
through a physical intervention, thus the combined
Attack Vector will be 1. On the other hand, consider-
ing that the thief (IDT) has O = 3 while the keylogger
(ES) has O = 4, their combined Opportunity is
further reduced to 2 (less than the minimum) since
the two single attackers must collude in order to be
effective.
Once a single value for every factor has been com-
puted, the procedure can continue as when consider-
ing a single attacker.
4.4 Usefulness of the Framework
Once the analyses have been performed, the outputs
of the framework can help properly tune the secu-
rity level of enrollment procedures: in many con-
texts, such as the financial one, different categories
of customers may have different needs to address; the
framework can help determine which solutions are
worth adopting depending on the risk level one is will-
ing to accept. Moreover, the possibility to analyse a
protocol by changing the set of security mitigations
implemented allows a what-if analysis, thus helping
designers understand the risks and analyse the effects
of mitigations on their protocols.
The framework can be used to model and anal-
yse the security and risk of any enrollment procedure,
since the set of actions involved (i.e., the specifica-
tion language) and the capabilities of the attackers
on the associated identification factors (i.e., the threat
model) can be fully customised and adapted (if nec-
essary) to the considered scenario.
In addition, the security and risk analysis prob-
lems that we have defined in Definitions 1 and 2 are
decidable. In the former case, the security analysis
problem considers finite sets, thus it is always possi-
ble to analyse all the involved attackers and combi-
nations of them to identify those violating the secu-
rity goal (if any). As for the latter case, instead, the
risk analysis problem takes in input the list of suc-
cessful attackers from the security analysis and eval-
uates the risk for each of them, thus working on finite
sets again. These considerations make the framework
suitable for implementation in an automated tool (cur-
rently an ongoing activity) to automatically assess the
security of enrollment procedures.
5 MITIGATIONS
By both analysing state-of-the-art solutions and elab-
orating some ideas ourselves, we have identified a list
of security mitigations that can be applied to enroll-
ment procedures involving eDocuments (as per Sec-
tion 3); these mitigations are displayed in Table 3
together with the effects they have on the attackers
belonging to AT T (and thus on their capabilities in
C). To better understand them, it is important to no-
tice that depending on how the interaction with the
eDocument happens – we can distinguish two scenar-
ios:
MRTD scenario: the applicant uses its eID card
and allows the app to interact with it by scanning
the MRZ (see Section 2);
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
227
Table 3: List of the identified mitigations and their effects.
# Mitigations Effects on attackers Source
M1
Provide the app with a root detection library, in
order to prevent the use of the app on rooted de-
vices.
ES, MA (makes the detection easier in case this
attacker forces a root of the device)
From (OWASP, 2020)
(§ 8.1).
M2
Leverage code obfuscation methodologies to pre-
vent attackers from fully understanding the source
code of the app (at least the most sensitive func-
tions).
MA (decrease the chance of understanding the
code, and thus to alter the app)
From (OWASP, 2020)
(§ 8.9).
M3
Implement, in the IAS ECC scenario, a binding
between the challenge and the certificate used to
establish a mutual TLS channel.
MA (only in the IAS ECC scenario: prevent the at-
tacker from injecting a stolen response in the legitimate
app)
Originally used in
OAuth 2.0 to bind ac-
cess tokens (Campbell
et al., 2020).
M4
Implement, in the MRTD scenario, a set of checks
on the authenticity of the provided data.
MA (only in the MRTD scenario: prevent the at-
tacker from sending fake information)
M5
Require users to capture a selfie at that moment,
preventing the upload of existent files.
SE £ (cannot reuse files that have been deceivingly
obtained by the user)
From many state-of-
the-art solutions (see
Section 7).
M6
Force the user to capture the selfie from the front
camera.
SS £ (cannot capture pictures or video of people in
the proximity)
From many state-of-
the-art solutions (see
Section 7).
M7
Request users to capture the selfie also in the IAS
ECC scenario.
IDT, ES £ (cannot obtain the selfie of the user)
SS, SE (have less probability of obtaining the selfie
of the user)
M8
Implement liveness detection libraries to detect
the misuse of static or modified pictures.
SE £ (cannot inject fake pictures on the legitimate
app)
SS (decrease the chance of obtaining pictures or
video of people in the proximity)
NB: MA is not affected as they can bypass the liveness
checks
From most state-of-
the-art solutions (Ju-
mio, 2020a; iDenfy,
2020b).
M9
Implement techniques to link the selfie captured
during the procedure to the ongoing session, thus
requiring the users to insert some fresh informa-
tion.
MA (cannot misuse already existent pictures or self-
ies that have already been used in other enrollment op-
erations)
From some state-of-
the-art solutions (Cas-
sidy, 2018).
£ = attacker made ineffective = attacker’s risk decreased
IAS ECC scenario: the applicant uses its eID
card and allows the app to interact with it by pro-
viding the PIN.
The proposed mitigations are fully compliant with
the requirements in Section 3, with particular re-
gard to R1: even those that are performed server-
side (e.g., M8 and M9) can indeed be implemented
through fully automated machine-learning algorithms
that require human intervention only when certain se-
curity thresholds cannot be achieved.
Security designers do not necessarily need to im-
plement all the proposed mitigations, as they can se-
lect those fitting best in their use cases. Of course, im-
plementing only a subset of the proposed mitigations
could affect the security and risk level of the proce-
dure.
In addition to the description in Table 3, it is worth
providing further details on some less-trivial mitiga-
tions, so as to better understand and thus manage to
implement them in a proper way.
M3. To prevent the use of a stolen response by
unauthorized parties, we suggest implementing a pro-
cedure similar to the one described for OAuth in RFC
8705 (Campbell et al., 2020). The mutual TLS chan-
nel that gets established does not aim to perform
client authentication (self-signed certificates do not
provide any guarantee), yet the binding of the app’s
self-signed certificate to the authentication challenge.
This way, as only the app knows the private key cor-
responding to its self-signed certificate, the server can
be sure that the response presented by a client con-
tains a challenge that was previously issued to the
same client, and thus that it has not been stolen or
tampered with. This procedure involves the following
additional steps:
1. before starting the enrollment procedure with the
eID card, the app must generate a key pair with a
self-signed certificate on the user’s mobile device;
2. the app establishes a mutual TLS channel with the
server using the generated key pair and certificate;
3. the server generates a challenge, associates the
SECRYPT 2021 - 18th International Conference on Security and Cryptography
228
client certificate with it and sends the certificate-
bound challenge to the app;
4. the app interacts with the eID card to sign the chal-
lenge. The response is sent to the server after es-
tablishing a mutual TLS channel;
5. the server verifies the response and checks that the
certificate used to establish the mutual TLS chan-
nel is the same that is bound to the challenge.
M4. In Section 2, we described how people’s per-
sonal data can be verified server-side in the IAS
ECC scenario through a challenge-response protocol.
Moreover, we stated that the MRTD scenario pro-
vides guarantees on the extracted information only
for in-person procedures, with physical readers ver-
ifying the Document Security Object (SOD, detailed
in (ICAO, 2015)). However, as briefly explained in
M4, we would need a way to trust the information ex-
tracted even during remote identification operations.
To this end, we could rely on the SOD in a different
way: since this object provides a digital signature of
all the eDocument’s data groups, the app could send
it along with the extracted data, so to allow the server
to perform the proper verifications on the authenticity
and integrity of the received data. In this way, MA at-
tackers trying to violate SG by sending fake informa-
tion without even interacting with a real eDocument
are prevented, since they would need a SOD that they
cannot provide since those personal data are fictional
and not extracted from an existing eDocument.
On the other hand, MA attackers that interact with
an eDocument through MRTD can also read the SOD:
for this reason, although the server can check the au-
thenticity of the data (i.e., the Validation part of the
enrollment flow, as detailed in Section 2), it cannot
verify that they really belong to the applicant (i.e., the
Verification part).
M7. As the picture contained in the eID card is
not included in the X509 certificate, M7 requires the
reader to access both the IAS ECC and MRTD ap-
plications (to retrieve the X509 certificate and the
owner’s picture, respectively). To do this, the MRZ
value must be inferred by following a specific proce-
dure (ICAO, 2015). Once this value has been derived,
the reader can mutually authenticate with the eID card
and access the picture as in the MRTD scenario.
M9. The most promising techniques to implement
this mitigation are the following:
S1. Require the user to write an alphanumeric
code suggested by the app on a piece of paper,
and take a selfie with it; the code aims to enable
the server to perform some verifications on the
link with the ongoing operation. Specifically, the
server should verify that: the user’s face from the
selfie matches the one from the eDocument’s pic-
ture, the code written on the paper is correct, and
the selfie has been sent within a certain interval.
S2. Require the user to capture a video of her-
self while reading some words aloud; these words
would be suggested by the app to enable the server
to perform some verifications on the link between
the words and the ongoing operation. Specifically,
the server should verify that: the user’s face from
the video-selfie matches the one from the eDocu-
ment’s picture, the words read aloud are correct,
and the video-selfie has been sent within a certain
interval.
6 FRAMEWORK APPLICATION
In this section, we apply the framework described in
the previous sections to a concrete enrollment pro-
cedure satisfying the requirements in Section 3 and
specifically based on eDocuments. We have per-
formed this work in collaboration with the Italian Fin-
Tech startup CherryChain
2
, by focusing on a realistic
use case inspired to their research and innovation ac-
tivities: one of the main projects they are developing
aims at realising a unique platform to connect service
providers (e.g., energy suppliers) and consumers, thus
handling the whole contracting lifecycle; the first step
of this project is represented by the enrollment.
After describing the enrollment procedure that we
have designed for this use case (Section 6.1), we per-
form the security and risk analysis on the designed
procedure (Sections 6.2 and 6.3).
6.1 Specification Language
In Section 5, we have identified the two possible sce-
narios when dealing with eDocuments; each of them
can be associated with an enrollment flow that we now
describe by detailing all the actions involved. Regard-
less of the scenario, each flow begins with some com-
mon actions: first, the user is requested to agree with
the privacy policy and terms of service (¯). Then,
she has to choose the type of document she would like
to use (in this case, eID card) and whether she wants
to provide the PIN or not (), thus being redirected
to the IAS ECC or MRTD scenario, respectively. The
next actions, instead, are strictly connected with the
considered flow:
2
https://www.cherrychain.it/
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
229
MRT D : ¯; ; ê(;); (æ); ¡; @; ;
IAS ECC : ¯; ; ÷(PIN); (æ); ¥; ¡; @; ;
6.1.1 Flow for the MRTD Scenario
After the common actions, the user is required to scan
the MRZ of her eID card through the device’s camera
(ê(;)) and read this document through the device’s
NFC capabilities ( (æ)). The residence address can
be extracted from the eID card, so the user only has to
confirm the correctness of all the data or modify them
in case, e.g., the residence address has changed (¡).
After providing and verifying both the email address
(@) and the phone number (), the user is required
to capture a selfie (
); the flow is then completed.
6.1.2 Flow for the IAS ECC Scenario
After the common actions, the user is required to in-
sert the PIN (÷(PIN)) and read her eID card through
her mobile device’s NFC capabilities ( (æ)). She is
then asked to manually insert the residence address
(¥) not present in the X509 certificate, thus can-
not be extracted from the eID card in this scenario
and confirm the correctness of all the data (¡). Once
provided and verified both the email address (@) and
the phone number (), the user is required to capture
a selfie (
); the flow is then completed.
6.2 Security Analysis
In this section, we perform a security analysis of the
enrollment procedure described in Section 6.1. Ac-
cording to the considered flow, we can identify a list
of identification factors involved:
MRTD: the eID card (æ) and the selfie ()
that is later compared to the eID card’s picture
SG
MRT D
=
{
æ,
}
IAS ECC: the eID card (æ), its PIN and the
selfie () that is later compared to the eID card’s
picture SG
IAS_ECC
=
{
æ, PIN,
}
To perform the security analysis, we have to un-
derstand how each attacker in AT T affects the iden-
tification factors; such capabilities (represented by C)
are displayed in Table 4 through the padlock notation
described in Section 4.2.
6.2.1 Analysis for the MRTD Scenario
The flow belonging to this scenario relies on the eID
card (
æ) and the selfie () as identification factors,
thus we have to focus only on these two columns in
Table 4; according to them, the only minimal subset
that manages to compromise the protocol is
{
MA
}
. In
Table 4: Attackers’ capabilities.
Attackers
Identification Factors
æ PIN
IDT
ES
SS
SE
MA
= safe = compromised
= indirectly compromised
fact, although M4 and M9 considerably affect this at-
tacker (by enabling the server to verify the integrity
of the received data and the connection between the
selfie and the ongoing enrollment operation, respec-
tively), MA still manages to bypass them: in the for-
mer case, by getting the victim’s personal data and
SOD by interacting with her eID card through the ma-
licious app on her device, then reusing these data on
his own device; in the latter case, by maliciously in-
teracting with the victim and deceiving her into pro-
viding the fresh selfie as required. Moreover, MA also
manages to bypass M8 by pretending to be sending a
fresh selfie.
Instead, some mitigations are completely effective
against some other attackers: by preventing the reuse
of selfies obtained through user’s deception, M5 and
M8 make SE ineffective; therefore, these mitigations
thwart the combination
{
IDT , SE
}
that would other-
wise be able to obtain both the eID card and the selfie.
Similarly, M6 and M8 defeat SS by preventing it from
taking pictures or videos of users in proximity; there-
fore, also the combination
{
IDT , SS
}
which would
be otherwise able to obtain the eID card and capture
an unauthorized picture of the user as if it was a user’s
selfie – is not effective.
6.2.2 Analysis for the IAS ECC Scenario
The flow belonging to this scenario relies on the eID
card (æ), the PIN and the selfie () as identification
factors, thus we have to focus on all the columns in
Table 4; according to them, even this scenario can
be violated only by the minimal subset composed of
{
MA
}
. Regarding the mitigations, M3 prevents
{
MA
}
from injecting a stolen signed challenge in legitimate
apps, since the TLS binding requires the app that gets
the challenge to be the same as the app that sends
the response back. However, MA could be success-
ful anyway when a signed challenge is captured by a
malicious app in the user’s smartphone and then in-
jected into another malicious app under the control of
the attacker. In this case, being both apps in control of
the attacker, MA could use the same key pair to estab-
lish the mutual TLS channel, thus managing to enroll
SECRYPT 2021 - 18th International Conference on Security and Cryptography
230
using information belonging to the victim. In the flow
connected with this scenario, MA can also bypass M8
by pretending to be sending a fresh selfie.
Instead, M7 in combination with M5, M6 and
M8 meets our expectations and makes
{
IDT , ES
}
,
{
IDT , SS
}
and
{
IDT , SE
}
ineffective. In fact, while
obtaining the eID card and its PIN was generally
enough, M7 also requires a user’s selfie.
It is important to notice that the selfie would not
be mandatory in this scenario: in fact, we could lever-
age the definition of derived credential (Grassi et al.,
2017b), which is a credential issued on the basis of an
already available credential that had previously been
issued after identity proofing, in order not to dupli-
cate this process. In this case, the possession of an
eID card and the knowledge of the related PIN would
prove the control of an authenticator (i.e., the eID card
itself) that has been previously issued by a trusted en-
tity (a Government office). Therefore, the Verifica-
tion phase described as part (iii) in Section 2 could
be fulfilled by simply inserting the eID card’s PIN in-
stead of providing additional elements such as a selfie.
Nevertheless, the considerations above prove that the
request for a user’s selfie has indeed relevant conse-
quences and significantly helps improve security.
6.3 Risk Analysis
The results of the risk analysis (in Table 5) highlight
that MA is connected with a Medium risk in both sce-
narios. By comparing the risk values to those of MA
with no mitigation implemented (Table 6), we can
notice the considerable effectiveness of the proposed
mitigations: beyond making most of the successful at-
tackers ineffective (see Section 6.2) they even reduce
the risk of MA, though without completely thwarting
it. For the sake of clarity, the risk values positively af-
fected by the implemented mitigations are highlighted
in bold in Table 5. Further details on the values as-
signed to the likelihood and impact factors in Tables 5
and 6 are detailed on the complementary website
3
.
As for M9, additional considerations are worthy:
during the risk analysis, no difference has been de-
tected between the effects of S1 and S2 on the overall
risk, since in both cases MA manages to interact with
the victim to obtain everything he needs. Specifically:
in case of S1, MA has to deceive the victim into pro-
viding the attacker with a selfie of her, together with
the code connected with the attacker’s malicious en-
rollment attempt written on a piece of paper; in case
of S2, MA has to deceive the victim into providing the
attacker with a video-selfie of her, in which she also
reads the words connected with the attacker’s mali-
3
https://st.fbk.eu/complementary/SECRYPT2021
cious enrollment attempt. Therefore, the choice above
can be made depending on other factors such as us-
ability, technical difficulty and/or server load. Nev-
ertheless, no matter which of the above techniques is
adopted, the overall risk is Medium in both scenarios.
In general, all the proposed mitigations significantly
reflect on the attackers’ likelihood, while having little
effect on the impact. Since the likelihood is already
Low, affecting the impact (which would be a quite
complex operation) would be the only way to further
decrease the overall risk.
7 STATE-OF-THE-ART
SOLUTIONS
Many works dealing with security methodologies ap-
plied to authentication protocols can be found in the
academic literature, either from a formal (Jacomme
and Kremer, 2021; Fett et al., 2016; Feng et al., 2021)
or an implementation perspective (Sudhodanan et al.,
2016; Yang et al., 2016). However, the same cannot
be said for enrollment: to the best of our knowledge,
no previous security methodology for such protocols
has been developed so far. Therefore, our state-of-the-
art review has focused on governmental (Section 7.1)
and commercial (Section 7.2) enrollment solutions,
which we have taken into consideration during both
the design of the flows and the identification of the
mitigations.
7.1 Governmental Solution: SPID
SPID (Sistema Pubblico di Identità Digitale, Public
Digital Identity System) (AgID, 2020) is the Italian
digital identity to access Public Administration’s on-
line services in a frictionless and secure way. Due
to its sensitiveness, SPID can be released only by
accredited Identity Providers; each of them may set
up their own procedures to properly identify users
(e.g., physical identification, remote identification
through webcam or digital signature). Given the focus
of this work, to extract basic actions and mitigations
to consider we have further studied the procedures re-
lying on eDocuments as identification means (e.g., the
Poste Italiane solution (Poste Italiane, 2020)).
7.2 Commercial Solutions
Many companies provide commercial solutions to
perform fully-remote identity verification, such as
(Authenteq, 2020), (HooYu, 2020), (iDenfy, 2020a),
(Jumio, 2020b), (Onfido, 2020), (Thales Group,
2020), (Veriff, 2020), and many other. Unfortunately,
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
231
Table 5: Risk analysis of the enrollment procedure with all the mitigations implemented.
Scenario Attackers Likelihood Impact Risk
TD O AV UI SA Overall AS AD Overall
MRTD
{
MA
}
3 1 7 1 2 2.80 Low 8 7 7.50 High Medium
IAS ECC
{
MA
}
2 1 7 1 2 2.60 Low 8 5 6.50 High Medium
Table 6: Risk analysis of the enrollment procedure without mitigations, considering only MA.
Scenario Attackers Likelihood Impact Risk
TD O AV UI SA Overall AS AD Overall
MRTD
{
MA
}
6 9 7 7 6 7.00 High 9 8 8.50 High Critical
IAS ECC
{
MA
}
3 2 7 1 4 3.40 Medium 8 6 7.00 High High
companies rarely release details about their proce-
dures, considering the commercial nature of their
business. For these reasons, since we are not their
customers, we can only refer to the available docu-
mentation that can be found online. The basic pro-
cedure shared between the various solutions (with
slight changes) usually requires the applicant to take
a picture of her document (either electronic or non-
electronic) and then a selfie, so that the photo ex-
tracted from the former can be compared to the lat-
ter. After collecting the needed information, some
checks are performed server-side; among these, the
most common is the liveness check that we have im-
plemented in our procedure as well.
Besides a basic procedure, some solutions also al-
low full customisation according to the customers’
needs: in case of (Veriff, 2020), for instance, by
adding or removing steps that the applicant must
perform (e.g., video-selfie or selfie with the eDocu-
ment) and by enabling or disabling checks to perform
(e.g., device verification, biometric comparison, live-
ness check and automated data extraction).
7.2.1 FinTech Solutions
To comply with Know Your Customer (KYC) guide-
lines, online banks have to identify prospective cus-
tomers with a high degree of assurance before let
them open a bank account. According to (Built for
Mars, 2020), some institutes still perform verifica-
tions manually, hence increasing the customers’ wait-
ing before their account is activated. Anyway, the
majority of the analysed banks leverage enrollment
procedures that are mostly performed in an automatic
way; manual intervention can still be needed, but only
in some cases (exceptions or over-threshold risks). To
this end, some institutes implement commercial solu-
tions (e.g., (Monzo, 2020) relies on Jumio and (Rev-
olut, 2020) on Onfido), while some other implement
custom enrollment procedures.
8 CONCLUSIONS
This paper has focused on enrollment procedures, that
are gaining more and more popularity since they al-
low a solid identification of the applicants. Given
the sensitive contexts in which these operations are
placed, they must provide a high level of assurance
on the applicants’ identities. Therefore, we have pre-
sented a framework for security and risk analysis of
any enrollment procedure; this framework is com-
posed of: (i) a specification language to provide a
clear and graphical description of enrollment proto-
cols; (ii) a security analysis methodology to obtain a
list of the attackers that are able to compromise the
protocols; (iii) a risk analysis methodology to sort the
successful attackers according to their severity, thus
enabling a prioritization of them. The outputs of the
framework can be used to adjust the security level
of enrollment procedures, also allowing to perform
what-if analyses by changing the set of mitigations
to consider and verifying the effects on security.
In the context of a collaboration with the Italian
FinTech startup CherryChain, we have applied the
proposed framework to fully-remote solutions rely-
ing on eDocuments as identity evidence. The collabo-
ration with CherryChain was extremely important to
both parties: on the one hand, it contextualized our
work in a practical use case; on the other hand, our
framework allowed CherryChain to verify the secu-
rity of the protocols they were designing, also iden-
tifying the mitigations to implement after discussing
their benefits in terms of security and feasibility.
Starting from this work, we plan to enrich the
specification language that has been used in this pa-
per to support a wider range of enrollment proce-
dures, even based on different requirements. More-
over, to simplify and automatize the analysis process,
we are currently formalising the proposed framework
through formal definitions and pseudocodes that can
be easily implemented within an automatic tool. Fi-
SECRYPT 2021 - 18th International Conference on Security and Cryptography
232
nally, it would be interesting to extend our work by
taking inspiration from some of the considerations
presented in a report (ENISA, 2021) released by the
European Union Agency for Cybersecurity (ENISA)
after this work was already completed.
ACKNOWLEDGEMENTS
The research has been partly supported by Cher-
ryChain S.r.l. in the context of a research and inno-
vation project funded by the Autonomous Province of
Trento, and by IPZS (Poligrafico e Zecca dello Stato
Italiano, the Italian Government Printing Office and
Mint) within the joint laboratory DigiMat Lab.
REFERENCES
AgID (2020). SPID. https://www.spid.gov.it/.
Authenteq (2020). Identity Verification Without Compro-
mise. https://authenteq.com/.
Built for Mars (2020). Opening an account. https://
builtformars.co.uk/banks/opening/.
Campbell, B., Bradley, J., Sakimura, N., and Lodderstedt, T.
(2020). RFC 8705. https://tools.ietf.org/html/rfc8705.
Cassidy, N. (2018). Onfido launches the market’s
most robust video liveness detection. https:
//onfido.com/resources/blog/news-onfido-launches-
the-market-s-most-robust-liveness-detection.
Cooper, D., Santesson, S., Farrell, S., Boeyen, S., Housley,
R., and Polk, W. (2008). RFC 5280. https://tools.ietf.
org/html/rfc5280.html.
ENISA (2021). Remote ID Proofing. https:
//www.enisa.europa.eu/publications/enisa-report-
remote-id-proofing.
Feng, H., Li, H., Pan, X., and Zhao, Z. (2021). A For-
mal Analysis of the FIDO UAF Protocol. In Network
and Distributed System Security Symposium, NDSS
2021. https://www.ndss-symposium.org/ndss-paper/
a-formal-analysis-of-the-fido-uaf-protocol/.
Fett, D., Küsters, R., and Schmitz, G. (2016). A Compre-
hensive Formal Security Analysis of OAuth 2.0. In
Proceedings of the 2016 ACM SIGSAC Conference
on Computer and Communications Security, CCS ’16.
https://doi.org/10.1145/2976749.2978385.
GIXEL (2009). European Card for e-Services and National
e-ID Applications: Identification Authentication Sig-
nature European Citizen Card (IAS ECC).
Grassi, P. A. and Fenton, J. L. (2017). Digital Identity
Guidelines: Enrollment and Identity Proofing. SP
800-63A. https://doi.org/10.6028/NIST.SP.800-63a.
Grassi, P. A., Fenton, J. L., Newton, E. M., Perlner, R. A.,
Regenscheid, A. R., Burr, W. E., and Richer, J. P.
(2017a). Digital Identity Guidelines: Authentication
and Lifecycle Management. NIST Special Publication
800-63B. https://doi.org/10.6028/NIST.SP.800-63b.
Grassi, P. A., Garcia, M. E., and Fenton, J. L. (2017b).
Digital Identity Guidelines. SP 800-63-3. https:
//doi.org/10.6028/NIST.SP.800-63-3.
HooYu (2020). Identify. https://www.hooyu.com/h/hooyu-
identify/.
ICAO (2015). MRTD. Doc 9303. https://www.icao.int/
publications/pages/publication.aspx?docnum=9303.
iDenfy (2020a). Identity Verification Service. https://www.
idenfy.com/.
iDenfy (2020b). Liveness Detection to prevent Spoofing At-
tack. https://www.idenfy.com/blog/spoofing-attack-
prevention/.
Jacomme, C. and Kremer, S. (2021). An Extensive For-
mal Analysis of Multi-Factor Authentication Proto-
cols. ACM Trans. Priv. Secur., 24(2). https://doi.org/
10.1145/3440712.
Jumio (2020a). Certified Liveness Detection. https://www.
jumio.com/technology/live-detection/.
Jumio (2020b). Jumio. https://www.jumio.com/.
Monzo (2020). Monzo. https://monzo.com/.
NIST (2017). Digital Identity Guidelines document suite.
SP 800-63. https://pages.nist.gov/800-63-3/.
Onfido (2020). User onboarding. https://onfido.com/use-
cases/user-onboarding/.
OWASP (2018). OWASP Risk Rating Methodol-
ogy. https://owasp.org/www-community/OWASP_
Risk_Rating_Methodology.
OWASP (2020). Mobile Application Secu-
rity Verification Standard. Version 1.2.
https://github.com/OWASP/owasp-masvs/releases/
download/v1.2/OWASP_MASVS-v1.2-en.pdf.
Pernpruner, M., Carbone, R., Ranise, S., and Sciarretta, G.
(2020). The Good, the Bad and the (Not So) Ugly of
Out-of-Band Authentication with eID Cards and Push
Notifications: Design, Formal and Risk Analysis. In
Proceedings of the Tenth ACM Conference on Data
and Application Security and Privacy, CODASPY
’20. https://doi.org/10.1145/3374664.3375727.
Poste Italiane (2020). PosteID abilitato a SPID. https://
posteid.poste.it/.
Revolut (2020). Get more from your money. https://www.
revolut.com/.
Sudhodanan, A., Armando, A., Carbone, R., and Com-
pagna, L. (2016). Attack Patterns for Black-Box
Security Testing of Multi-Party Web Applications. In
Network and Distributed System Security Symposium,
NDSS 2016. https://www.ndss-symposium.org/wp-
content/uploads/2017/09/attack-patterns-black-box-
security-testing-multi-party-web-applications.pdf.
Thales Group (2020). ID Verification Suite. https:
//www.thalesgroup.com/en/markets/digital-identity-
and-security/government/identity-verification.
Veriff (2020). Veriff. https://www.veriff.com/.
Yang, R., Lau, W. C., and Liu, T. (2016). Sign-
ing Into One Billion Mobile App Accounts
Effortlessly with OAuth2.0. In Black Hat Eu-
rope 2016. https://www.blackhat.com/docs/eu-
16/materials/eu-16-Yang-Signing-Into-Billion-
Mobile-Apps-Effortlessly-With-OAuth20-wp.pdf.
A Framework for Security and Risk Analysis of Enrollment Procedures: Application to Fully-remote Solutions based on eDocuments
233