Alex Smirnoff,

Oleg Makarov,
Glanc, Ltd

Addressing user
authentication challenges
for cryptocurrency exchange:

NEW THREAT MODEL AND IMPLEMENTATION CONSIDERATIONS

Objective
The threat model for cryptocurrency exchange is somewhat unique even for finance
applications; two essential techniques that make anti-fraud efforts fruitful in traditional
fintech, are KYC and the capability to dispute a suspicious operation. In the
cryptocurrency exchange, these may turn totally non-reliable or even nonexistent since
crypto currency operations are generally anonymous and non-reversible. Thus, we need
more rigorous (and preventive) security as compared to traditional banks and payment
systems1. The traditional instruments we typically have at hand are obviously insufficient.
We do not discard KYC completely, and we try to work around non-reversibility wherever
possible. However, the purpose of this paper is to explore ways to surpass the intrinsic
limitations of traditional methods, because the opposite approach of making case by
case improvements is already getting enough public attention. A mindless combination
of different authentication and recovery methods may merge into a cascade of failures I
instead of increasing redundancy and reliability; thus we need a systemic approach.

Executive summary
There is no reliable way to achieve a reasonable level of security with passwords only II in
the context of protecting high-value resources, yet passwords are the necessary first layer
of protection (given the server-side credential storage is implemented in a secure manner III).
To maintain a proper balance of security and usability, it is necessary to be flexible in
protection requirements, in accordance with the monetary value at risk.

IF THERE IS NO SUBSTANTIAL VALUE, THERE IS NO NEED TO COMPLICATE

THE USER’S EXPERIENCE WITH UNNECESSARY VERIFICATION STEPS.

Two-factor authentication should be mandatory, preferably in the form of hardware OTP

tokens or hardware certificate tokens. The latter are notably more efficient if equipped with
an on-device screen and a pin pad for digital signing of individual operations by the user
within the device. For the period before these tokens are available for the user, we propose
a transitional solution in the form of a one-time code sheet.
SMS and email are generally not considered as means of secure communication, mostly
due to account ownership issues and social engineering problems; yet they could be used
to deliver additional information about current operations to the customer. Email security
could be significantly improved if end to end encryption is available (which usually is not
due to the low adoption rate of the technology). SMS as a second factor may also be used
as a temporary low-security option for newly registered accounts with a relatively small
amount of funds which certainly do not exceed the cost of even most low-tech attacks and
should be appropriately discarded afterward.

1 It would be unwise to completely ignore the NIST 800-63 Digital Identity Guidelines, yet those are even less relevant
for our case as a high-level approach; however, it is still a valuable reference material for many technical aspects.

2
 It is advised to implement additional authorization to perform high-risk operations
(substantial funds transfer or changing access credentials), and in a case when a risky
method is used for credentials management (like, changing the primary email address tied
to an account while no secondary secure authorization method is defined).

IT IS ADVISABLE TO LIMIT POSSIBLE OPERATIONS WITH NEW CREDENTIALS

FOR A GUARD PERIOD WHEN ALL CHANGES COULD BE REVERSED, AND
OPERATIONS THAT CANNOT BE REVERSED ARE TEMPORARILY PROHIBITED.

Scope and prerequisites

We consider 3 types of authentication/authorization to be within the scope of this memo:
1. “MAIN” user authentication, as at the beginning of user session;
2. “RECOVERY” methods to reinstate access if main credentials are lost or unavailable;
3. “ADDITIONAL” authorization for the pre-authenticated user to perform
high-risk operations (like sending substantial funds out of the system or changing
access credentials);
also, one special scenario which does not fall into one of these categories: initial user
registration process (see the corresponding section for details).
We intentionally leave out of scope:
• In-depth analysis of user behavior to identify suspicious and automated activity;
• Endpoint and communication protection details (beyond common considerations);
• Session management, persistent cookies and related questions (those matters are
well-covered by OWASP recommendations IV);
• Mobile applications and associated app-only security features like using a protected
key store provided by mobile OS, device pinning, etc.;
• Out of band user verification specifics (it is likely to be an external service).

3
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Online brute force attacks. Behavior analysis, rate Some users would
Password
Main

COMPLEXITY: LOW limiters, password always be susceptible

strength meters. to this kind of attack.
COST: LOW Password strength
meters are unreliable and
RISK: HIGH contradictory because
there is no objective metric.
Also, need to consider
availability issues to make
it sure legit users won’t be
blocked out of the system.

Offline brute force attacks. Protecting credential Though “hashes leak” is

COMPLEXITY: HIGH storage, “hard” a low probability situation
password-based key which also indicates
COST: VARIES derivation functions. some serious breach,
consequences may be
RISK: MEDIUM catastrophic at scale, so it
is necessary to implement
adequate protection.

Compromised endpoints Out of scope – We cannot deny users

(offline attacks) user’s responsibility. the possibility to save
note: this part applies to passwords, and attempts
saved passwords only, to do so would do more
everything else is related to harm than good.
the session management
(cookies and cache data).
COMPLEXITY: MEDIUM

COST: VARIES

RISK: HIGH

Compromised endpoints Out of scope –

(online, a.k.a. synchronous user’s responsibility.
attacks: keyloggers and
direct session intervention).
COMPLEXITY: MEDIUM

COST: VARIES

RISK: MEDIUM

4
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Lost/stolen credentials on Out of scope – Stronger passwords
Password
Main

physical media. user’s responsibility. are typically harder to

COMPLEXITY: LOW memorize, and certainly,
some of those are to be
COST: VARIES written down or saved in
text files on random places.
RISK: MEDIUM

Phishing attacks. Limit legitimate interactions If there is some action

COMPLEXITY: LOW that would involve following required from the user side,
links sent by email; easily we should never ask to
COST: VARIES recognizable URLs, “follow this link and log in
EV certificates.1 V to complete the operation”.
RISK: HIGH All messages of this type
should be delivered in-
system (like, log in from the
previously bookmarked link
or a manually typed URL
and then do whatever
is needed).

Compromised endpoints Make sure recovery codes Once information is sent

Recovery codes
Recovery

(offline attacks). page would not be stored in out to the user, you cannot
COMPLEXITY: MEDIUM browser cache; download control it.
should be on demand and
COST: VARIES not unsolicited.

RISK: MEDIUM

Compromised endpoints Reduce attack window by Also, see “credentials

(online attacks). making it clear and visible management” section
COMPLEXITY: MEDIUM which recovery codes user below.
has enabled and when.
COST: VARIES

RISK: MEDIUM

1 Unfortunately, email marketing tools like MailChimp made “nonsense URLs” a new norm (the HTML letters make
it even easier, and do not require a 3rd-party – in HTML you may plainly hide a URL behind any word (looks like
everything is designed to confuse a user)). And EV certificates are obviously dying out. Anyway, there is no systemic
solution to the phishing problem and never was.

5
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Lost/stolen credentials on Out of scope – More likely lost than stolen.
Recovery codes
Recovery

physical media. user’s responsibility.

COMPLEXITY: LOW

COST: VARIES

RISK: MEDIUM

Cryptographically insecure Use strong cryptographic Set to low risk because it is

(predictable) recovery pseudo-random function or easy to mitigate.
codes. true random numbers.
COMPLEXITY: MEDIUM Implementation review.

COST: VARIES

RISK: LOW

Online brute force attacks, Recovery questions are Truly random huge answers
Control questions
Recovery

social engineering attacks, evil, don’t use them. would be not any worse
data mining attacks. than passwords, but the
COMPLEXITY: LOW existence of recovery
questions as we know it
COST: LOW certainly encourages users
to dangerous behavior.
RISK: HIGH

Identity theft, third If done properly, identifying Might be very undesirable

Physical identification (local)
Recovery

party risks. a user by “real world” for many users because

COMPLEXITY: VARIES credentials is reliable of privacy issues (and
enough, but generally availability too, people
COST: VARIES requires trusted third party travel). Also, it is a “slow”
like a notary. method and requires a
RISK: MEDIUM trusted global partner.

6
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Identity theft, third Works better if a user has If no “deposited” identity
Physical identification (remote)
Recovery

party risks. “deposited” his credentials exists, making “web

COMPLEXITY: VARIES (say, live photo with ID camera quality” fake
document) in advance. documents is trivial. Also,
COST: VARIES the recent shift in the video
processing technology
RISK: MEDIUM makes “online video” even
less reliable.

Compromised endpoints. User’s responsibility, A typical user accesses

Recovery links sent via email
Recovery

COMPLEXITY: MEDIUM however, make sure email accounts from

recovery links a) expire multiple devices, which
COST: VARIES after a brief period; contributes to the risk.
b) cannot be reused.
RISK: HIGH

In-transit interception End to end email One specific factor to

(transit server or encryption, make sure consider here is DLP
communication channel recovery links a) expire and content inspection
compromised). after a brief period; systems, including “lawful
COMPLEXITY: HIGH b) cannot be reused. interception”.

COST: HIGH

RISK: MEDIUM

Mail server compromised. End to end email Today most users rely on
COMPLEXITY: HIGH encryption, make sure cloud email providers which
recovery links a) expire are usually reasonably
COST: HIGH after a brief period; secure, yet we should
b) cannot be reused. include this risk in the
RISK: MEDIUM threat model because
environment may vary.

7
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Mail account compromised End to end email End to end email
Recovery links sent via email
Recovery

(scenarios include: hacked encryption, make sure encryption is not

accounts; accounts on recovery links a) expire widely adopted.
expired domains; loss of after a brief period;
ownership when using b) cannot be reused.
work account; cloud
account takeover via social
engineering).
COMPLEXITY: LOW

COST: LOW

RISK: HIGH

Mail account ownership End to end email End to end email

expiration encryption. encryption is not widely
(expiring domains, work adopted.
accounts, etc.).
COMPLEXITY: LOW

COST: LOW

RISK: MEDIUM

Cryptographically insecure Implementation review. Set to low risk because it is

(predictable) recovery links. easy to mitigate.
COMPLEXITY: MEDIUM

COST: VARIES

RISK: LOW

Vulnerabilities in the Implementation review,

server-side recovery source code audit.
application.
COMPLEXITY: VARIES

COST: LOW

RISK: MEDIUM

8
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Compromised mobile Out of scope – This type of risk could be
SMS codes and links
Main, Recovery, Additional

devices. user’s responsibility. easily mitigated by user.

COMPLEXITY: VARIES

COST: MEDIUM

RISK: MEDIUM

Phone number ownership Could be partially mitigated SIM card fingerprinting

loss (expiring contracts). if a partner MNO supports is available only if local
COMPLEXITY: LOW ICCID or IMSI fingerprinting. MNO supports it and is
unsuitable for worldwide
COST: LOW operations.

RISK: MEDIUM

Phone number hijacking May be a partially mitigated SIM card fingerprinting

(social engineering if a partner MNO supports is available only if local
or access to the MNO ICCID or IMSI fingerprinting. MNO supports it and is
facilities). unsuitable for worldwide
COMPLEXITY: MEDIUM operations.

COST: LOW

RISK: HIGH

SMS interception via SS7 Out of scope.

signaling attacks.
COMPLEXITY: MEDIUM

COST: MEDIUM

RISK: MEDIUM

9
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
SMS interception via the Limiting lifespan of links Could be “low”, but actually
SMS codes and links
Main, Recovery, Additional

sending provider. and codes and proper happens.

COMPLEXITY: HIGH deactivation is vital to
make sure past links
COST: VARIES cannot be used or re-used.

RISK: MEDIUM

SMS interception via radio Out of scope.

networks (IMSI catchers).
COMPLEXITY: HIGH

COST: HIGH

RISK: LOW

Compromised endpoints Out of scope. Hardware OTP tokens

Hardware OTP tokens
Main, Additional

(online attacks). almost completely mitigate

COMPLEXITY: MEDIUM the possibility of offline
attacks.
COST: VARIES

RISK: MEDIUM

Lost or stolen tokens. Even if OTP is set, a

COMPLEXITY: VARIES password should still
be required; set a guard
COST: VARIES period after the password
recovery; use PIN-protected
RISK: MEDIUM tokens.

Key seed lifecycle Do not use tokens that See RSA SecurID breach
management issues. are pre-programmed with story.
COMPLEXITY: HIGH vendor keys.

COST: HIGH

RISK: LOW

10
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Server-side application Implementation review, Need to pay attention to
Hardware OTP tokens
Main, Additional

vulnerabilities. source code audit. server-side credentials

COMPLEXITY: VARIES storage, because current
OTP technologies require
COST: LOW key material that could
be reversed to token
RISK: MEDIUM seeds to be stored on the
authenticating side.

Compromised endpoints Out of scope. Software OTP tokens

Software OTP tokens
Main, Additional

(online attacks). transfer the offline attack

COMPLEXITY: MEDIUM risks to the authenticator
device.
COST: VARIES

RISK: MEDIUM

Compromised authenticator User education about

devices (offline attacks). mobile security (a software
COMPLEXITY: VARIES token is likely to reside on a
mobile phone).
COST: VARIES

RISK: HIGH

Lost or stolen authenticator On-device authentication Phones are more likely

devices. and encryption. Even if to be lost or stolen than
COMPLEXITY: VARIES OTP is set, a password dedicated authentication
should still be required; set tokens.
COST: VARIES a guard period after the
password recovery; use
RISK: HIGH PIN-protected tokens.

Server-side application Implementation review, Need to pay attention to

vulnerabilities. source code audit. server-side credentials
COMPLEXITY: VARIES storage, because current
OTP technologies require
COST: LOW key material that could
be reversed to token
RISK: MEDIUM seeds to be stored on the
authenticating side.

11
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Compromised endpoints Even if certificate is used, Using a certificate token
X.509 certificates (hardware storage)
Main, Additional

(online attacks). a password should still itself to sign operations is

COMPLEXITY: MEDIUM be required; set a guard the only reliable method
period after the password when the user’s computer
COST: VARIES recovery; use PIN-protected cannot be trusted.
tokens and tokens with However, it is costly and
RISK: MEDIUM on-device screens. complicated on the user’s
If a token device does not side (it requires token
have hardware buttons hardware with a display).
to authorize its operation,
it is essential to keep it
unplugged when not in use.

Lost or stolen tokens. Even if certificate is used,

COMPLEXITY: VARIES a password should still
be required; set a guard
COST: VARIES period after the password
recovery; use PIN-protected
RISK: MEDIUM tokens.

Server-side application Implementation review,

vulnerabilities. source code audit.
COMPLEXITY: VARIES

COST: LOW

RISK: MEDIUM

Compromised endpoints
X.509 certificates (software storage)
Main, Additional

(online attacks).
COMPLEXITY: MEDIUM

COST: VARIES

RISK: MEDIUM

12
1 — Type and description; 2 — Application Authentication methods comparison
Threats and attack
Mitigation Notes
1
2 characteristics
Compromised endpoints Passphrase protection of This risk is generally higher
X.509 certificates (software storage)
Main, Additional

(offline attacks). software certificate storage. than a risk of hardware

COMPLEXITY: MEDIUM tokens being stolen.

COST: VARIES

RISK: MEDIUM

Server-side application Implementation review,

vulnerabilities. source code audit.
COMPLEXITY: VARIES

COST: LOW

RISK: MEDIUM

Secure credentials storage

A thorough architecture review is required to determine that even if the main application is
compromised, the risk to credential storage is minimal (that applies not just to passwords,
for which the impact, given OTP use is to be mandatory, is mostly reputational, but to other
key material as well). For passwords it is always recommended to use yescrypt or similar
PBKDF implementation that addresses the attack path in multiple points, simultaneously
making it hard to retrieve hash data and to attack it once it is retrieved.

High-risk operations
It is advised to implement additional authorization to perform high-risk operations
(substantial funds transfer or changing access credentials), and in case a risky method is
used for credentials management (like changing the primary email address tied to account
while no secondary secure authorization method is defined).
Additional authorization may include methods mentioned in this document or anything
out of its scope, yet deemed feasible (obtaining a biometric voiceprint via a phone call,
for example).

13
Guard period and funds lock-in
In the context of resisting credit card fraud the following practice is very successful.
Funds that were deposited to the account via a credit card payment remain unavailable
for withdrawal / external transfers for a specific period, typically 5-10 days. We advise
extending this type of lock-in to all cases of account access when the account access
authenticity cannot be immediately verified, as described below in “Credentials
management scenarios” section.

Event notifications
Message delivery methods that were deemed insufficiently secure to be used to send one-
time codes for authentication or authorization purposes, like SMS and email, could still
be useful to keep the user aware of operations being performed on his account. However,
the best known way is to sign individual operations with hardware electronic signature
token that is also capable of displaying detailed information about what exactly the signed
operation is. When altering notification methods, the change should not be in effect
immediately: the old method should remain active (and clearly marked for retention in all
notifications) at least for 5 days, giving the user an option to contact the customer support
in case of suspected fraud.

Credentials management scenarios

A user account may be in one of the following states:
1. INSECURE – after initial registration, before the user has any secure credentials
associated with account: there is no substantial amount of funds at user’s disposal
(less than $100-200, as for current estimation we derived from the probable
corresponding attack cost) and no strong authentication (see secure credentials
below) methods are defined. This state allows “relaxed” access recovery procedure if
there is nothing of value to protect. Before any funds are transferred to the system,
it is required for the user to define one or more type of secure credentials to maintain
further account access (NB: scenarios may vary depending on business requirements,
and this part requires special attention to ensure there are no loopholes
or race conditions).
2. SECURE – the user has secure credentials associated with the account. It is
mandatory for all accounts exceeding the threshold limit mentioned above (Google
Authenticator, Authly and alike on mobile phones, most probably). For accounts with
the balance exceeding some higher limit, use of a hardware OTP token should
be encouraged.
3. INACTIVE/DISABLED.
An account can also be optionally (if it does not contradict the privacy requirements)
verified if the user provides a government issued id, which could later be used for access
recovery (see corresponding table entries for caveats). Guard period could also be imposed
on certain operations (either funds transfer or both funds transfer and credentials change).

14
 We consider credentials (and user contact methods that may be relevant for access
recovery) to be one of three types:
1. SECURE – any type of credentials that could be reasonably presumed to be secure
if appropriately handled on the client side. Secure credentials include any type of
a second factor, such as software or hardware tokens, or one-time codes used for
access recovery; PGP-encrypted email; it does not include reusable passwords,
because passwords are susceptible to numerous threats and once compromised could
be used for further access indefinitely. Caveat: the term “secure” should not be taken
literally as an absolute characteristic; there are attack scenarios (see table above) that
are still possible when using this type of credentials, just the residual risk is lower.
2. INSECURE – any type of credentials that either likely to be partially outside
of user’s control (plaintext email, phone numbers) or could be easily reused once
compromised (passwords).
3. INTRINSIC – user identification which is presumed to be protected from arbitrary
manipulation by “real world” constraints; this includes government-issued ID’s,
biometric data and so on. Using this type of identification may have privacy drawbacks,
and requires a trusted, established process which is likely to be slow. This method
requires human support intervention, and thus we do not cover it in details. It is
recommended, however, that the guard period would be in effect after using this
method as well.

GENERAL RULE: YOU NEED ONE “REGULAR” (“INSECURE”) AND AT LEAST

ONE OF “SECURE” CREDENTIALS SIMULTANEOUSLY TO CHANGE A THIRD ONE
(REGARDLESS IF IT IS “SECURE” OR NOT) INSTANTLY BYPASSING THE
“GUARD PERIOD” IMPOSED OTHERWISE.

Some kind of “secure” credentials is required to be set up to get full account access.

WE STRONGLY ADVISE AGAINST TREATING A COMBINATION OF INSECURE

CREDENTIALS AS A SECURE METHOD.

Say, two-part verification code consisting of a link sent via e-mail and a PIN sent to a
phone is not reliable, because if the phone is compromised, it is very likely that email
would be taken over as well. The security is not cumulative! Pay attention to the real world
dependencies between the auth methods (outside of the secure system in question).
For example:
If you have presented a password and an OTP token (the same combination which is
enough to log in), you may change email or phone.
If you have email access and OTP (or backup codes), you may request instant password
recovery (that’s why you probably do not want to keep both in a single place, like
authenticator app and mobile email account being on the same smartphone);
If you have email access but no OTP nor backup codes, then the password recovery should
be burdened with a guard period.

15
 If you have lost both access to your email and your password, but have your OTP code, you
need to contact the customer support first, and the account should be, after appropriate
identity verification, limited with guard period.
All those operations should be accompanied by email notification and SMS notification if a
phone number is set.
Adding recovery methods, and changing authentication methods should be possible with
additional authorization only. Customers are encouraged to routinely review authentication
and recovery methods that are enabled for their accounts. (it is important to note that
typical user interfaces are not review-friendly)

Transitional one-time codes

This document defines several occasions when it is mandatory for the customer to use
secure credentials. It might be reasonable to provide hardware OTP tokens to high-value
customers for free; and the rest is encouraged to use their own hardware or software
tokens, the latter being supported by all major mobile platforms. It is possible, however,
that a user does not have the mobile device he wishes to use as OTP token at hand at the
moment of the registration, and delivering the hardware token takes time, so that won’t
be available right away. We need to provide temporary access to those users until they get
their permanent authentication device, and we cannot rely on insecure credentials even
for a brief period.
As a transitional solution, we suggest using a code sheet, consisting of 10 numbered
6-digit one-time codes that could be used before a permanent method is established, along
with the recommendation to print it or write it down (not to save on a computer! Though
we probably should expect it to happen). Codes are relatively short because they are to be
typed frequently and possibly to be written down by hand (see scenario examples below
for details about backup access codes); however, they are totally unsuitable for long-
term use for obvious reasons; one of those is that the sheet is likely to reside in a wallet
where it wears out quickly. Making a photo might be a not that bad idea too, but the user
should be warned about risks associated with cloud photo backups. While using this code
sheet customer gets continuously warned that he should switch to authenticator app (or a
hardware token if he was issued with one, depending on account balance); and that after
using the 10th code it won’t be possible to log in anymore. The sessions after the 5th code
should only provide the ability to add a new OTP token.

The new user registration process

Current recommendations:
• If immediate email access verification is not possible in a direct way on early stage
(though should be completed afterward), the email address is to be entered twice to
avoid typos (as it is usually done with passwords).
• Upon the completion of the registration, a user is advised to enable full account access
by defining a secure second factor. If he has an OTP token, he may wish to use it as the
second factor. If he wishes to use his present mobile device the necessary software

16
 should be provided. Otherwise, he must write down a recovery code and keep it in a
safe place until he acquires the hardware he needs. The current session is considered
to be secure and provides full access to credential management until the user logs out
or is logged out automatically.
• If a user makes an initial payment with a credit card, a guard period is imposed over
the funds regardless of the amount.
• If a user made an initial payment with a credit card, the same credit card could be
used to reinstate account access after a possible registration glitch if it has 3DSecure
support or equivalent.

Acknowledgements
We would like to express our gratitude to Roman Akopov (Bixtrim), Eugene Panferov
(independent expert and valuable author at ITHipster), Oleksandr Nikitin (Lynx Capital)
and Andrey Kovalev (ThreatMetrix) for critical discussion and suggestions on this paper.

17
Annex A: Example scenarios
Let’s consider 3 different threat models for accounts, depending on the funds on the
account balance: 1) if peak total amount is less than $150, we consider the overall
situation to be of low risk, since any targeted attack would be impractical; such accounts
may lack secure credentials, and that’s fine. 2) if account value is in the range $150-
$10000, then using secure software-based OTP is recommended. 3) for accounts
exceeding $10000 balance, it might be practical to provide a hardware OTP token. If we
cannot immediately provide the user with an appropriate OTP token, a transitional
one-time code system (see the corresponding section above) should be used as a
temporary solution.
Scenarios that imply a change of primary notification/contact methods (phone, email)
should not rely on the availability of old contact method since the motive for a change
could be that user no longer has control over it because it has been deactivated,
compromised or expired. We also removed email and phone reset scenarios if the
password is not known: such a situation is exceptional enough to be handled by customer
support on a per-case basis.
Assumptions:
• There is no user id, a user is identified by email. All scenarios may be easily adjusted to
use a separate user id if needed, or to use social network profile as a source of contact
data (email and phone).
• Phone number with SMS capability is a mandatory attribute (which could be somewhat
disputable for a privacy-focused service).

1. NEW USER REGISTRATION

Initial state: a user is not logged in, an account does not exist;
Final state: an account is registered, a user is logged in;
Entry fields: phone number, email (two times), password (two times), captcha
(optional), SMS code.
a) Log in (see below “Insecure login” and “Secure login”) or register – a user is asked
to provide (as a minimum) phone number (verification code is sent and required
to be entered), email address (twice to avoid typos) and password (twice to avoid
typos). Email and phone should be verified later (by visiting a link sent by email
and by entering a verification code sent via SMS), but this affects nothing except
frequent reminders to do so.
b) Optionally, add OTP token as defined below. A user may skip this step.

18
 2. INSECURE LOGIN
Initial state: a user is not logged in, an account exists in the insecure state;
Final state: an account is insecure, a user is logged in;
Entry fields: email, password, SMS code, captcha (optional).
A user enters email, password and SMS code and is permitted to enter an
insecure session. SMS and email notifications are being sent about login attempt
(Unsuccessful, mandatory. Successful, optional. Mind possible flood situations, so
event aggregation is required). If any other active user session exists, it is terminated
forcibly upon successful login.

3. SECURE LOGIN
Initial state: a user is not logged in, an account exists in secure state;
Final state: an account is secure, a user is logged in;
Entry fields: email, password, second factor, captcha (optional).
A user enters login, email and second factor and gets full account access. SMS
and email cannot be used as a second factor. SMS and email notifications are
mandatory for unsuccessful logins and optional for successful ones (same caveats
as previous case). If any other active user session exists, it is terminated forcibly
on successful login.

4. PASSWORD CHANGE
Initial state: a user is logged in, an account exists and may be secure or not;
Final state: a user is logged in, the password is changed;
Entry fields: old password, a new password (2 times), second factor (optional).
For insecure accounts, SMS code may be used as a second factor, or the second factor
could be omitted completely (depends on business requirements). If a second factor
exists and the account is secure, it should be used for this operation. SMS and email
notifications are issued in both cases.

5. PASSWORD RESET
Initial state: a user is not logged in, an account exists and may be secure or not,
a user has email access, the password is not known to a user;
Entry fields: email, captcha; after the link is opened in the browser – a second factor
and a new password (2 times);
Final state: a user is logged in, the password is changed.

19
 For insecure accounts SMS code is used as a second factor; for secure accounts,
it should be an OTP token or a recovery code. A link is sent by email (not SMS!)
combined with the second factor is enough to perform the change.

NB: THE LINK SENT BY EMAIL SHOULD CONTAIN NO ENTRY FORMS

FOR SENSITIVE INFORMATION, JUST A BUTTON TO CONFIRM ACTION
AND, POSSIBLY, CAPTCHA!

6. EMAIL ADDRESS CHANGE/RESET

Initial state: a user is logged in, an account exists and may be secure or not;
Final state: a user is logged in, the email address is changed;
Entry fields: new email, password, second factor, SMS code.
If the account is not secure, an SMS code is enough to perform the change, if it is,
both SMS code and second factor are required. Password should be re-entered as well
to avoid session hijack situations. Email notifications are being sent to both new and
old address for next 5 days, and a link sent to the old email address could be used to
either confirm the operation or temporarily block account access (support intervention
is required to resolve the dispute afterward). A link is also sent to the new email to
verify it is correct.
Attacks to consider:
a) An attacker in temporary possession of user’s account credentials, possibly
including OTP, tries to take over the account quickly.
b) An attacker takes over the victim’s email (and, possibly, phone), and it cannot
be easily recovered; thus user needs to change the email address.
Disabling the account temporary to prevent abuse is an acceptable inconvenience
in all those cases.

7. PHONE NUMBER CHANGE/RESET

Initial state: a user is logged in, an account exists and may be secure or not;
Final state: a user is logged in, the phone is changed;
Entry fields: new phone, second factor (if one exists), SMS code to confirm the change.
This scenario is almost similar to one for the email address change/reset. If a
second factor is not defined and the account is not secure, a link sent by email and
a confirmation code are enough to perform the change. If second factor exists and
account is secure, it should be used for this operation. The password should be re-
entered as well to avoid session hijack situations. Phone notifications are being sent
to both new and old phone for next 5 days, and a link sent to the old phone could be
used to either confirm the operation or temporarily block the account access (support
intervention is required to resolve the dispute afterward). A link is also sent to the new
email to verify it is correct.

20
 Attacks to consider:
a) An attacker in temporary possession of user’s account credentials, possibly
including OTP, tries to take over the account quickly.
b) An attacker takes over the victim’s phone (and, possibly, email) and it cannot
be easily recovered; thus the user needs to change the phone number.
Disabling the account temporary to prevent abuse is an acceptable inconvenience
in all those cases.

8. ADDING AN OTP TOKEN

Initial state: a user is logged in, an account exists and may be secure or not;
Final state: a user is logged in, a second factor is added, the account status
is assigned as secure;
Entry fields: password, second factor to test.
Notifications are sent via email and SMS. We have two types of backup access codes
that could be used in absence of “proper” OTP authenticator:
a) Temporary access codes are generated if the user is forced to use OTP, but has
no authenticator at hand or does not wish one right now. Such codes are short
(6-digits) because the user needs to type it in frequently, the total number is
limited to 10, and after 5 are already used a notification should appear with a
reminder to switch to “proper” OTP soon. When the user adds “proper” OTP token,
all temporary codes are to be instantly disabled.
b) Permanent backup access codes are used to restore account access if OTP token
is lost, stolen, damaged, malfunctioning or compromised. Such codes may be
considerably longer and alphanumeric, and creation of those codes is considered
a mandatory part of OTP token provisioning.

There are some possible concerns if it is a good idea to allow more than one OTP token
at a time. Since we cannot prevent a user from using one seed with multiple tokens, it is
better to give him a possibility to do it more conveniently, having individual control over
authenticator devices. Since notifications are being sent out if a user adds a new OTP
device, it is unlikely that a new token could be added be a malicious actor without drawing
the user’s attention. Signing individual operations (using “interactive” tokens) is to be
revised in future versions.

21
 9. REMOVING AN OTP TOKEN
Initial state: a user is logged in, an account exists and is secure;
Final state: a user is logged in, the second factor is removed, a user is prompted
to add a new OTP token as defined above;
Entry fields: password.
The user should be able to log in using a recovery code first. Notifications are being
sent via email and SMS. If a user has no OTP tokens left (other than recovery codes),
he is prompted to set up a new one (see “Adding OTP token”).

10. OTP TOKEN RESET IN THE ABSENCE OF BACKUP CODES

Initial state: a user is not logged in, an account exists and is secure;
Final state: a user is logged in, the second factor is removed, a user is prompted
to add a new OTP token as defined above, the account is being put on a 5-day
guard period;
Entry fields: email, password, SMS confirmation code.
A recovery link is sent by email and the second confirmation code is sent via SMS.
The guard period is still mandatory.

11. HANDLING INCOMPLETE CREDENTIALS CHANGE SCENARIOS

Until changes are fully confirmed (confirmation sent, the new code was entered, etc)
the new settings (if to be added) should not be in effect and old credentials (if to be
removed) should be treated as valid. If a user makes a purchase, but fails to complete
creation of secure credentials, the purchase (and the beginning of the corresponding
guard period, if applicable) should be deferred until the user successfully enables
secure credentials. During the guard period changing the phone number and the email
address should not be permitted.

12. EXCEEDING THE BALANCE THRESHOLD FOR THE INSECURE ACCOUNT

(AS MENTIONED ABOVE, $100-200)
If account balance exceeds the threshold for any reason, the user is directed to the
OTP token creation procedure which he cannot cancel (the account cannot be used
normally until an OTP is provisioned).

22
I Mohammad Ghasemisharif, Amrutha Ramesh, Stephen Checkoway, Chris Kanich, Jason Polakis
O Single Sign-Off, Where Art Thou? An Empirical Analysis of Single Sign-On Account Hijacking and Session
Management on the Web
https://www.cs.uic.edu/~polakis/papers/sso-usenix18.pdf

II Denise Ranghetti Pilar, Antonio Jaeger, Carlos F. A. Gomes, Lilian Milnitsky Stein
Passwords Usage and Human Memory Limitations: A Survey across Age and Educational Background
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0051067

III Solar Designer

yescrypt: large-scale password hashing
http://www.openwall.com/presentations/BSidesLjubljana2017-Yescrypt-Large-scale-Password-Hashing/

IV OWASP Session Management Cheat Sheet

https://www.owasp.org/index.php/Session_Management_Cheat_Sheet

V Troy Hunt
Extended Validation Certificates are Dead
https://www.troyhunt.com/extended-validation-certificates-are-dead/

VI Alex Smirnoff
A Hacker’s Guide to Not Get Hacked
http://ithipster.com/blog/72.html

+359878830030
[email protected]
facebook.com/glancltd
Varna, Bulgaria