Leaving this course?

Lesson 1 of 12

A customer is not an employee

Almost all authentication guidance is written for the workforce, and most of it does not transfer. The differences are structural rather than a matter of degree.

What you have with an employee and not with a customer:

  • A device you manage. You chose it, you configure it, you can require a particular authenticator on it.
  • A training obligation. They can be made to attend, and told what a phishing attempt looks like.
  • A help desk with a relationship. The person calling is in your directory, has a manager, and has a staff number.
  • The ability to compel. An employee who refuses the authentication method does not keep their job. A customer who refuses it goes to a competitor, and that is a rational response rather than a compliance failure.
  • One identity per person. Customers share accounts with spouses, use a relative's phone, and hand the device to a son to operate.

And one thing you have with a customer that you do not with an employee: a regulator with an opinion about the authentication itself. In payments this is not advisory. Strong customer authentication is specified in law, down to the number of failed attempts permitted and the amounts below which you may skip it, and lessons seven to nine are about that.

So this course is about authenticating customers — in payments, in banking and in any service where the person on the other side chose to be there and can leave. The workforce side of the subject, including privileged access, joiner-mover-leaver discipline and machine identity, is covered in the identity and access management course, and this one does not repeat it.

Why "after onboarding" is the frame. An enormous amount of effort goes into proving who someone is at account opening — document checks, biometric comparison, liveness, data verification. That work establishes identity once. Every subsequent interaction is a different question: is this the same person as the one who was proved? Authentication answers that question, every day, for years, and it is usually given a fraction of the attention and budget that onboarding received.

One consequence of that last point deserves naming. With an employee, the account holder and the operator are the same person. With customers they often are not: a spouse who knows the password, a son who operates the banking application for an elderly parent, a small business whose owner has given the credentials to a bookkeeper. Authentication will say the right credential was used, truthfully, while the person at the keyboard is someone else — and in most of those cases nothing criminal is happening. A design that treats every such session as fraud will generate alerts nobody can work.

The failure mode this course exists to prevent. An institution proves a customer to a high standard at onboarding, protects the account with a password and an SMS code, and is surprised when accounts are taken over. Nothing about the onboarding was wrong. The chain simply has a second link, and the next lesson is about why it is usually the weaker one.

Lesson 2 of 12

The chain: proofing, authentication, and the weaker link

NIST's digital identity guidelines separate three things that institutions routinely blur, and the separation is the most useful idea in this course.

Three claims, measured independently.

  • Identity assurance level (IAL) is about enrolment: how strongly was this person proofed when the account was created? Remote proofing of the kind the eKYC courses describe — validated evidence, a chip read, a biometric comparison against that evidence, liveness — is the territory of IAL2. The top level adds a trained representative interacting with the applicant in an attended session and the collection of at least one biometric characteristic.
  • Authentication assurance level (AAL) is about returning: how strong is the sign-in that connects today's session to that enrolled account?
  • Federation assurance level covers the case where one party vouches to another.

Why the separation matters. The chain is as strong as its weakest link, and the links are independent. Proof a customer to the highest standard available, then protect the account with a guessable password, and an attacker does not need to defeat your document checks, your liveness detection or your biometric matcher. They log in.

That sentence is worth reading twice, because the budget usually goes the other way. Onboarding is a project with a vendor, a business case and a launch. Authentication is a setting.

What the levels require, briefly. At AAL2, the guidelines require that verifiers offer at least one phishing-resistant authentication option. At AAL3, a phishing-resistant authenticator with a non-exportable authentication key is required — which has a direct consequence for passkeys that lesson four explains.

The practical question for a service. Not "what AAL are we at?" but "which operations require which level?" Reading a balance, changing an address, adding a payee and sending money to a new account are different acts and do not need the same proof. A service with one authentication event at the front door and nothing afterwards has decided that every action is equally consequential, which it is not.

And the two directions this gets wrong. Too little, and account takeover produces a payment the customer never authorised. Too much, and the customer abandons a legitimate transaction, calls the contact centre, or stops using the channel — at which point the cost is real and lands on the business rather than on the fraud line.

What the rest of the course does. Lessons three to six cover the authenticators themselves and what each is actually worth. Seven to nine cover what payments regulation requires, which is the part no workforce guidance will tell you. Ten to twelve cover the three things that decide whether the design survives contact with real customers: when to ask again, how recovery is handled, and what happens to the person who cannot use your chosen method.

Lesson 3 of 12

Passkeys, and what phishing resistance actually means

A passkey is a public key credential created by an authenticator and used through the web authentication standard. The private key stays with the authenticator; the service holds only the public key. There is no shared secret, so there is nothing for a customer to repeat to someone who asks.

The mechanism that defeats phishing, stated precisely. The standard binds a credential to the party it was created for: "A public key credential can only be used for authentication with the same entity (as identified by RP ID) it was registered with."

Read what that means operationally. A customer taken to a convincing replica of your site cannot use their passkey there, because the browser will not offer a credential registered for your domain to a different domain. The customer does not have to notice the deception. The protocol does it, which is the whole point — every control that depends on a person spotting a wrong address eventually fails, because people are busy.

This is the property that nothing else in common use has. A one-time code can be read out to a caller. A push approval can be accepted under pressure. A password can be typed into anything.

Replay is handled separately. Each authentication uses a fresh challenge from the server, and the standard sets a floor: "In order to prevent replay attacks, the challenges MUST contain enough entropy to make guessing them infeasible. Challenges SHOULD therefore be at least 16 bytes long." A captured response is of no use for a second attempt.

What a passkey does not do. Three things worth being clear about, because they are where real deployments go wrong:

  • It does not prove who the person is, only that the authenticator registered to this account was used and unlocked. The strength of that link back to a real person comes from the enrolment in lesson two, not from the protocol.
  • It does not protect against a compromised device. An attacker with control of an unlocked phone has the passkey.
  • It does not remove the need for recovery, and recovery is where the attacker will go instead. Lesson eleven is about that, and it is the single most important lesson in this course for anyone deploying passkeys.

Registration is the moment that matters. A passkey registered during a session the attacker controls is the attacker's passkey, and it is phishing-resistant for them. So the step-up before registering a new credential has to be at least as strong as the credential being created — which is easy to get wrong when the feature is launched as a convenience and bolted onto whatever the sign-in already was.

Lesson 4 of 12

Synced or device-bound, and why the difference is not cosmetic

Passkeys come in two forms, and vendors tend to describe both with the same word. The distinction has consequences for assurance, for recovery and for what you are allowed to claim.

The two forms. The standard's terms: "A backup eligible public key credential source is referred to as a multi-device credential whereas one that is not backup eligible is referred to as a single-device credential." In ordinary language, a synced passkey is copied by a cloud service across a person's devices; a device-bound passkey never leaves the one authenticator that made it. The FIDO Alliance uses that same split, noting that security keys have housed device-bound credentials since 2019.

The flag is permanent, and that is useful. "Backup eligibility is a credential property and is permanent for a given public key credential source", and its value "is set during authenticatorMakeCredential operation and MUST NOT change." So a service can tell at registration which kind it has been given, and can decide what to allow on that basis. It cannot be changed later, which means the decision is yours at registration and nobody else's afterwards.

Why syncing helps the customer. A new phone works immediately. The customer does not lose access when a device is lost, which removes the most common reason a person falls back to the recovery process — and recovery is the weakest part of most designs.

And what it costs in assurance. Syncing requires the private key to be exportable from the authenticator into the provider's sync service, which breaks the assumption behind the highest assurance level. NIST is explicit: "Since syncable authenticators (described in Appendix B) require the private key to be exportable, syncable authenticators SHALL NOT be used at AAL3."

The security of a synced credential therefore depends on an account you do not control — the customer's platform or password manager account. NIST addresses this directly, requiring that access to keys in the sync service "SHALL be protected by AAL2-equivalent MFA to preserve the integrity of the authentication protocols using the synced keys", and that access controls ensure "only the authenticated user can access their authentication keys in the sync fabric".

Note what that means in practice: your customer's platform account is now part of your authentication chain, and you cannot audit it.

How to decide. For most consumer services, synced passkeys are the right default: the assurance is sufficient, and the reduction in recovery events is a real security gain rather than only a convenience. Where a single operation genuinely warrants the top level — a high-value transfer, a change of registered details — require a device-bound credential for that operation specifically, rather than for everything.

And say which you are relying on. An institution claiming a level of assurance it cannot support, because it never checked the flag it was sent, has made a statement it cannot defend.

Lesson 5 of 12

Passwords for customers, and the rules that changed

The password guidance most organisations enforce was written to compensate for the absence of anything better, and the standards have now moved decisively against it.

What NIST SP 800-63B-4 requires. Three provisions, and the second and third reverse what most services still do:

  • Length: "Verifiers and CSPs SHALL require passwords that are used as a single-factor authentication mechanism to be a minimum of 15 characters in length." Shorter is permitted where the password is only one factor of several, but not below eight.
  • Composition: "Verifiers and CSPs SHALL NOT impose other composition rules (e.g., requiring mixtures of different character types) for passwords."
  • Rotation: "Verifiers and CSPs SHALL NOT require subscribers to change passwords periodically."

Both prohibitions are shall not, not need not. A service still forcing a symbol and a quarterly change is not merely dated; it is doing something the standard forbids.

Why the reversal. Composition rules do not produce unpredictable passwords; they produce a capital letter at the front and a digit and an exclamation mark at the end, which attackers model. Forced rotation produces a predictable increment on the previous one, and it trains people to write passwords down. Both rules made the secret worse while feeling rigorous.

And the customer-specific problem with the 15-character rule. For employees you can deploy a password manager and require it. You cannot require anything of a customer. A fifteen-character minimum imposed on a consumer population without one produces three outcomes: reuse of a password from another site, a note in a phone's contacts, or abandonment at registration.

So read the length requirement together with the one in lesson two — that a verifier must offer at least one phishing-resistant option at AAL2. The way to meet both is not a longer password rule. It is to make the password a fallback rather than the mechanism, and to put the passkey in front of it.

What to do instead, in order:

  • Offer a passkey as the primary route, with the password retained for customers who cannot or will not use one.
  • Allow long passphrases, accept every character including spaces and emoji, and do not truncate silently. A service that accepts a long password and quietly uses the first twelve characters has a security problem its customers cannot see.
  • Check new passwords against known-breached lists rather than against composition rules. That is the check that actually reduces risk.
  • Stop expiring them, and rotate on evidence of compromise instead.
  • Never send one, never display one, never log one, and never let a contact-centre agent see or set one for a customer.

The workforce side of password policy is covered in the identity and access management course; what is specific here is that your population cannot be trained, equipped or compelled.

Lesson 6 of 12

SMS, the PSTN and the SIM swap

The one-time code sent by text message is the most widely deployed second factor in consumer services and the weakest one in common use. It is worth being precise about why, because "SMS is insecure" is too blunt to design with.

What the standard says. NIST SP 800-63B-4 treats it as a category of its own: "At the time of publication of these guidelines, there is one restricted authenticator: the use of the PSTN for out-of-band authentication." Restricted, not prohibited — permitted with conditions and a documented risk acceptance, which is a deliberate position given how many services depend on it.

The guidance also says what to do before relying on it: "Verifiers SHOULD consider risk indicators (e.g., device swap, SIM change, number porting, other abnormal behavior) before using the PSTN to deliver an out-of-band authentication secret." That is an actionable requirement. Operators in many markets expose a signal for how recently a number was ported or a SIM re-issued, and a service that sends a code to a number that changed hands this morning has the information and chose not to use it.

The three distinct weaknesses.

  • The code is phishable. It is a secret the customer can be persuaded to read to someone. Nothing about the delivery channel changes that, which is why an SMS code is not phishing-resistant at any length.
  • The number can be taken. A SIM swap moves the number to an attacker's SIM. The FCC's order describes the consequence plainly: it "gives the bad actor control of the customer's phone number so that the bad actor receives the text messages and phone calls intended for the victim."
  • Delivery is not guaranteed. Messages are delayed, dropped and intercepted, which produces a support cost and trains customers to re-request codes — behaviour an attacker can exploit.

What regulators have done about the first link. The FCC's 2023 order revised its rules "to require wireless providers to adopt secure methods of authenticating a customer before redirecting a customer's phone number to a new device or provider", and specified that authentication methods "shall not rely on readily available biographical information, account information, recent payment information, or call detail information", with a review "not less than annually".

And one provision worth copying into your own contact centre. The same rule requires that providers "establish safeguards and processes so that employees who receive inbound customer communications are unable to access CPNI in the course of that customer interaction until after the customer has been properly authenticated." An agent who can see the customer's details before authenticating them will read them out to confirm, and a caller who hears their own address read back has just been authenticated by the agent in reverse.

The honest position. If you must use SMS, use it as one factor among several, check the porting and swap signals, never use it for recovery of a stronger credential, and have a plan to move off it.

Lesson 7 of 12

Strong customer authentication: what the law requires

In the European payments regime, authentication is not an engineering choice. Strong customer authentication is defined in the directive and specified in a delegated regulation, and the detail is unusually prescriptive.

The definition. PSD2 defines it as "an authentication based on the use of two or more elements categorised as knowledge ... possession ... and inherence ... that are independent, in that the breach of one does not compromise the reliability of the others."

Independence is the operative word and the most commonly failed requirement. A password typed into an application on a phone, confirmed by a push notification to the same application on the same phone, is two steps on one device. If the device is compromised, both are.

What the technical standards require. The delegated regulation sets out, among other things:

  • Two or more elements from different categories, resulting "in the generation of an authentication code".
  • A cap on retries: "the number of failed authentication attempts that can take place consecutively ... shall not exceed five within a given period of time".
  • A session limit: "the maximum time without activity by the payer after being authenticated for accessing its payment account online shall not exceed 5 minutes."
  • Dynamic linking for remote payments, which is the provision worth understanding properly.

Dynamic linking, and why it defeats a whole class of attack. The regulation requires that "the authentication code generated is specific to the amount of the payment transaction and the payee agreed to by the payer", and that "any change to the amount or the payee results in the invalidation of the authentication code generated."

The consequence is that an attacker who obtains a valid authentication code cannot use it for a different payment. A code approving fifty shillings to a known payee cannot be replayed to approve fifty thousand to a new one. PSD2 itself states the requirement at the directive level: for remote electronic payments, providers must apply SCA "that includes elements which dynamically link the transaction to a specific amount and a specific payee."

This is why a customer-facing approval screen must display the amount and the payee, and why displaying "approve this transaction?" without them is not a cosmetic failure.

And the liability consequence, which concentrates the mind. PSD2: "Where the payer's payment service provider does not require strong customer authentication, the payer shall not bear any financial losses unless the payer has acted fraudulently." Ordinary payer liability for an unauthorised transaction is capped at fifty euro; where SCA was not applied, the payer bears nothing at all.

Where cards fit. For card payments made remotely, the authentication is usually carried by the EMV 3-D Secure protocol, which EMVCo describes as supporting SCA "by enabling the use of two-factor authentication" and in which consumers "must authenticate themselves using a challenge, such as a one-time passcode, knowledge-based questions, biometrics, or other methods". EMVCo recommends version 2.2 or higher for the full functionality. The protocol is the carrier; what it carries still has to satisfy everything above.

So the question "should we apply an exemption here?" is a question about who carries the loss, which is the subject of the next lesson. A proposal for a regulation to replace PSD2 was published in 2023 and has not been confirmed in the Official Journal, so the directive and its delegated regulation remain the operative law.

Lesson 8 of 12

The exemptions, and the arithmetic behind them

A rule requiring authentication on every payment would be unusable, so the technical standards provide exemptions. They are numeric, and knowing the numbers is most of the competence in this area.

Low-value remote payments. The exemption may be applied where "the amount of the remote electronic payment transaction does not exceed EUR 30", provided the cumulative amount since the last authentication "does not exceed EUR 100", or the count "does not exceed five consecutive individual remote electronic payment transactions."

Note the structure: a per-transaction ceiling and a running counter, either of which exhausts the exemption. A provider that implements the thirty-euro check and not the hundred-euro cumulative total is not compliant, and will discover it in an audit rather than in production, because nothing breaks.

Contactless at the point of sale. The same shape: "does not exceed EUR 50" per transaction, with a cumulative limit that "does not exceed EUR 150", or a count that "does not exceed five."

Transaction risk analysis, which is the interesting one. A provider may skip authentication on a higher-value payment if its own fraud rate is low enough, and the standards set out the trade explicitly in a table. At an exemption threshold value of EUR 500 the reference fraud rate is 0.01 per cent for card payments and 0.005 per cent for credit transfers; at EUR 250 it is 0.06 and 0.01; at EUR 100 it is 0.13 and 0.015.

Read that as the regulator pricing a privilege. Keep your fraud losses below a published figure and you may exempt larger payments; exceed it and the permission is withdrawn. It makes fraud performance a commercial asset rather than only a cost line.

The standards also name the real-time checks required to use it, including "abnormal spending or behavioural pattern of the payer", "unusual information about the payer's device/software access", "malware infection in any session of the authentication procedure", and "abnormal location of the payer".

Account access, and a number that changed. The original rule allowed access to account information without re-authentication for 90 days. That was amended: the period is now "more than 180 days have elapsed since the last time the payment service user accessed online the information specified in paragraph 1 and strong customer authentication was applied."

Anyone working from material written before that amendment will state 90 days, and will be wrong by a factor of two.

And one exemption that is mandatory rather than optional. The amending regulation provides that providers "shall not apply strong customer authentication where a payment service user is accessing its payment account online through an account information service provider" — with a carve-out permitting it where there are "objectively justified and duly evidenced reasons relating to unauthorised or fraudulent access". Shall not, by default. A provider adding friction to third-party access because it would rather customers used its own application has the rule the wrong way round.

Lesson 9 of 12

What actually counts as possession or inherence

The three categories sound self-evident and are not. The European Banking Authority published an opinion on the elements precisely because institutions were classifying things wrongly, and its findings are the most useful part of the regime for anyone designing a flow.

A code sent by text message is not the possession element. The opinion is specific: "In the case of an SMS, and as highlighted in Q&A 4039, the possession element 'would not be the SMS itself, but rather, typically, the SIM-card associated with the respective mobile number'."

That distinction decides what you are relying on. If the thing possessed is the SIM, then everything in lesson six applies: your possession element can be transferred to an attacker by a shop assistant at a mobile operator, and your control over that process is nil.

An application is only a possession element if it is bound to the device. The opinion: "an app or web browser that does not ensure a unique connection with a device would not be a compliant possession element." An application that can be installed on a second phone, or a web session that can be resumed anywhere after a login, demonstrates possession of nothing. The binding has to be to something that does not travel — a key in the device's secure hardware, which is what a device-bound passkey provides and a cookie does not.

Printed codes are out. The opinion states that "printed matrix cards or printed OTP lists that are designed to authenticate the PSU are not a compliant possession element for approaches currently observed in the market." Several institutions had treated a paper list as possession of an object. It is knowledge printed on paper.

And a caution about inherence. The opinion notes that the data points in a card authentication protocol "would not currently appear to constitute inherence elements". Inherence means something the person is — a biometric characteristic — not a behavioural signal collected during the session. Device fingerprints, typing rhythm and navigation patterns are valuable fraud signals and are not inherence in this sense. Classifying them as such produces a flow that looks like two factors and is one.

Push approvals, which sit awkwardly. The EBA has accepted that a push notification can evidence possession, provided the regulation's requirements on protecting possession elements are met, and that "It is for each payment service provider to prove these requirements are met." The practical weakness is not the mechanism but the human: a customer receiving an approval prompt while on the telephone to someone claiming to be the bank will approve it. Dynamic linking from lesson seven is the defence — the prompt must show the amount and the payee, so that approving the wrong thing requires ignoring what is on the screen.

And a standing caveat on the Q&As. They carry the EBA's own disclaimer that they reflect the provisions in force on their publication date and are not systematically reviewed afterwards. Check the date of any Q&A before relying on it.

Lesson 10 of 12

Step-up: deciding when to ask again

Authenticating once at the front door and never again treats every action in a session as equally consequential. Step-up authentication is the alternative: a baseline for entry, and a further challenge at the moment a particular operation is requested.

The pattern, and a standard for expressing it. RFC 9470 defines how a protected resource can tell a client that the authentication it holds is not sufficient, using an error code: "insufficient_user_authentication: The authentication event associated with the access token presented with the request does not meet the authentication requirements of the protected resource."

It carries two parameters that make the demand specific. The first bounds how stale the authentication may be: "max_age: This value indicates the allowable elapsed time in seconds since the last active authentication event associated with the access token." The second states what kind is wanted: "acr_values: A space-separated string listing the authentication context class reference values in order of preference."

The value of having this standardised is that the resource declares its requirement, rather than the sign-in screen guessing in advance what the session will be used for.

What should trigger a step-up. Two kinds of trigger, and a design needs both:

  • The operation. Adding a payee, changing registered contact details, raising a limit, registering a new authenticator, exporting data, or any payment above a threshold. These are known in advance and belong in configuration, not in a model.
  • The risk signal. A new device, a new location inconsistent with the last, a change of SIM or recently ported number, an access pattern that does not fit, or a session that has been idle. The regulation names its own list for the exemption in lesson eight, and the same signals serve here.

Time matters too, and there are numbers. NIST sets reauthentication bounds by level: at AAL2, "A definite reauthentication overall timeout SHALL be established, which SHOULD be no more than 24 hours at AAL2. The inactivity timeout SHOULD be no more than 1 hour." At AAL3 the overall timeout "SHALL be no more than 12 hours", and at AAL1 it is thirty days. In payments, the five-minute inactivity limit for online account access from lesson seven is a separate and much shorter constraint.

Two ways this goes wrong.

  • Stepping up on everything, which trains customers to approve prompts without reading them and makes the approval worthless at the moment it matters.
  • Stepping up with something weaker than the credential being protected. A passkey-protected account where a new payee can be added with an SMS code has an SMS-protected account with extra steps. The step-up must be at least as strong as what it guards, and the same applies to registering a new authenticator — the point from lesson three.

The test for a design. List the ten operations that would do the most damage if performed by somebody else, and write down what each currently requires. The gaps are usually obvious and nobody has made the list.

Lesson 11 of 12

Account recovery is the real attack surface

Every authentication design has a path for the customer who has lost their credential, and that path is the actual security of the account. An attacker will not attack the passkey. They will claim to have lost it.

Why this is the structural weakness. Recovery exists to let someone in who cannot authenticate, which is a description of what an attacker wants. And it is usually built by a different team, later, under pressure from support volumes, with the goal of reducing call handling time. The result is a side door protected by information rather than by possession.

Recovery that is weaker than the credential is the whole flaw. If a passkey-protected account can be recovered with a date of birth and the last three transactions, the account is protected by a date of birth and the last three transactions. Everything said about phishing resistance in lesson three applies only to the front door.

What the standard requires of recovery secrets. NIST bounds their validity tightly by channel: "Issued recovery codes SHALL be valid for at most: ... 10 minutes when sent via text message or voice ... 24 hours when sent to an email address", and where a code is pre-generated for the subscriber to keep, "The recovery code SHALL include at least 64 bits from an approved random bit generator."

Ten minutes for a voice or text channel is much shorter than most implementations allow, and the reason is in lesson six: that channel can be redirected.

What a defensible recovery process looks like.

  • Re-prove identity rather than re-confirm knowledge. The strongest recovery re-runs a proofing step — a document and liveness check of the kind used at onboarding — rather than asking questions whose answers are in a breach dump.
  • Delay, and notify. A recovery that completes instantly gives the customer no chance to object. One that takes an hour, with a notification to every channel on record including the one being changed, gives the real customer a chance to stop it.
  • Never let a change of contact details and a recovery happen in one session. An attacker who can change the email address and then recover to the new one needs nothing else. This single rule prevents a large share of real takeovers.
  • Require more than one factor of recovery evidence, not one.
  • Keep the contact centre out of it where you can. An agent under pressure, with the customer's details on screen, is the easiest route in — which is why the provision quoted in lesson six, keeping details hidden from the agent until the caller is authenticated, is worth copying.
  • Register a second authenticator at enrolment. The cheapest recovery is the one that never happens because the customer has another credential.

And measure it. How many recoveries completed, how many were later disputed, and how long each took. A recovery process whose dispute rate nobody knows is a control nobody has evaluated.

Lesson 12 of 12

The customer who cannot use your method

Every authentication design excludes somebody, and the exclusion is rarely visible to the people who designed it, because they all have recent smartphones.

Who is affected. A customer with no smartphone, or an older one that cannot hold a passkey. A customer sharing a phone with a household. A customer whose hands cannot complete a gesture, whose sight cannot read a code, or whose memory cannot hold a password. A customer in an area with no reliable mobile signal, for whom an SMS code arrives twenty minutes late or not at all. A customer whose fingerprints are worn by manual work, which the inclusion material in the eKYC courses documents.

For a bank, excluding these people is not an edge case. It is a group that includes the elderly, the poor, the disabled and the rural — and in many markets, a duty.

The standards require an alternative, in two places. On biometrics, NIST: "An alternative non-biometric authentication option SHALL always be provided to the subscriber." On the restricted channel, the same document requires that "verifiers SHALL ensure that alternative authenticator types are available to all subscribers". Both are shall.

And payments regulation does not prescribe the method. This is widely misunderstood, so it is worth quoting the EBA's answer on customers without a mobile phone directly: "PSD2 and the Commission Delegated Regulation (EU) 2018/389 do not prescribe a specific authentication approach that should be used for applying strong customer authentication." And: "It is left for payment service providers to choose how to authenticate their customers."

So "the regulation requires an app" is not true. The requirement is two independent elements from different categories. A hardware token, a card reader, or a code-generating device meets it without a smartphone, and the EBA notes that published lists of elements "include authentication elements that are not based on mobile phones and mobile applications."

Accessibility has a specific criterion worth knowing. WCAG 2.2 added success criterion 3.3.8, Accessible Authentication: "A cognitive function test (such as remembering a password or solving a puzzle) is not required for any step in an authentication process unless that step provides at least one of the following" — among them an "Alternative: Another authentication method that does not rely on a cognitive function test", or a "Mechanism ... available to assist the user in completing the cognitive function test."

Read plainly: a password field that blocks pasting fails this, because pasting from a password manager is the assisting mechanism. So does a puzzle with no alternative. A passkey passes it outright, which is an accessibility argument for passkeys that rarely gets made.

The design conclusion. Offer at least two routes that do not share a dependency, make the alternative discoverable rather than buried in a help article, and measure who uses it. If the alternative is used by nobody, either it is not needed or it cannot be found, and those are different problems with different fixes.

And the sentence this course ends on. Authentication that a customer cannot complete is not secure. It is unused, and the customer is on the telephone to an agent who will authenticate them with three questions from a breach dump.

Knowledge check

Fifty questions

Answer all fifty questions, then check your answers. You need 45 out of 50 to pass and receive a certificate. If you score less, you will see which answers were right and wrong, and then go through the course again before you retake the check. Your answers, progress and times are kept only in this browser.

Standards and frameworks covered

The published frameworks this course teaches toward. Saolix is not affiliated with, endorsed by or accredited by any of these bodies; naming a standard is a citation, not a claim about its author.

Sources

The official documents this course relies on. Laws and guidance change, so check the current version.

  1. Web Authentication: An API for accessing Public Key Credentials, Level 3 (W3C Recommendation, 25 August 2026) · World Wide Web Consortium (W3C)
  2. Passkeys: synced and device-bound credentials · FIDO Alliance
  3. SP 800-63B-4: Digital Identity Guidelines, Authentication and Authenticator Management (July 2025) · US National Institute of Standards and Technology (NIST)
  4. SP 800-63A-4: Digital Identity Guidelines, Identity Proofing and Enrollment (July 2025) · NIST
  5. Directive (EU) 2015/2366 (PSD2), Articles 4(30), 74 and 97 · Official Journal of the European Union, via EUR-Lex
  6. Commission Delegated Regulation (EU) 2018/389: regulatory technical standards for strong customer authentication · European Commission, via EUR-Lex
  7. Commission Delegated Regulation (EU) 2022/2360: the 180-day account access exemption and third-party access · European Commission, via EUR-Lex
  8. Opinion on the elements of strong customer authentication under PSD2 (EBA-Op-2019-06, 21 June 2019) · European Banking Authority
  9. Single Rulebook Q&A 2020_5325: alternative strong customer authentication for citizens without a mobile · European Banking Authority
  10. Single Rulebook Q&A 2019_4984: push-based authentication and SCA requirements · European Banking Authority
  11. EMV 3-D Secure · EMVCo
  12. RFC 9470: OAuth 2.0 Step Up Authentication Challenge Protocol (September 2023) · Internet Engineering Task Force (IETF)
  13. RFC 6238: TOTP, Time-Based One-Time Password Algorithm (May 2011) · IETF
  14. Web Content Accessibility Guidelines (WCAG) 2.2, success criterion 3.3.8 Accessible Authentication · W3C
  15. FCC 23-95: Protecting Consumers from SIM Swap and Port-Out Fraud (November 2023) · US Federal Communications Commission
  16. Guideline on Cybersecurity for Payment Service Providers (July 2019) · Central Bank of Kenya
  17. Data Protection Act, 2019 (No. 24 of 2019), sections 2 and 41 · Office of the Data Protection Commissioner, Kenya