What this course adds, and to whom
Two courses in this catalogue already introduce reusable digital identity. The decentralised identity course covers the model and what it means for customer due diligence (CDD, which much of the industry calls KYC); the eKYC deep-technology course ends with the signed-credential stack as the successor to inspecting documents. Neither tells you how to accept one.
This course is for the people who have to build or operate the verifier, or the issuer. It assumes you already accept the premise — an issuer signs a claim once, a holder keeps it, a verifier checks the signature — and it concerns itself with everything that premise leaves unanswered.
What is actually unanswered, and these are the twelve lessons:
- Which format. There are two in production use and they are not interchangeable, which surprises people who have read only the architecture documents.
- Which protocol. Issuance and presentation are specified separately, and a profile exists precisely because the specifications alone do not guarantee two implementations will interoperate.
- In person or over the internet. Different standards, different threat models, and one of the formats cannot be used for the first.
- Who do you trust. A signature from an issuer you should not trust verifies perfectly and means nothing. Trust moves; it does not vanish.
- Is the holder the subject. A valid credential proves what the issuer signed, not who is holding the phone.
- Is it still valid. Revocation is the hard part, as it always was, and checking it without letting the status service watch the holder is an unresolved tension.
- Can presentations be linked. The privacy properties people assume are present have to be designed in, and several are not yet deployable.
- What assurance can you claim. Two frameworks, different vocabularies, and one explicit statement about where a mobile licence sits as evidence.
- What happens when the phone is lost, which is the question every pilot discovers late.
- What is final and what is draft, which matters more here than in most fields because several load-bearing specifications are not finished.
- Where the law and the standards disagree, which they currently do in two documented places.
- And the deadline, which is not printed as a date anywhere and has to be computed.
One honest limitation, stated at the start. The normative text of the ISO standards in this field is behind a paywall. This course cites their published abstracts and catalogue entries, and takes its clause-level detail from the specifications that are open — the W3C recommendations, the IETF documents, the OpenID Foundation specifications, the EU legislation and the European Commission's reference framework. Where a fact would require quoting a paid standard, the course says what the standard covers rather than pretending to its wording.
Two formats, and they are not interchangeable
The single most consequential implementation decision, and the one most often made by accident.
The first is the ISO mobile document. ISO/IEC 18013-5 is the mobile driving licence standard, which "establishes interface specifications for the implementation of a driving licence in association with a mobile device". Its data model, usually called an mdoc, is a binary structure designed for a reader and a device to exchange directly, and it came out of the driving-licence world — which is why its first deployments are state licensing authorities.
Two scope notes from its own abstract are worth knowing, because they describe what you still have to build: out of scope are "how mDL holder consent to share data is obtained" and "requirements on storage of mDL data and mDL private keys."
The second is the selectively disclosable JSON token. SD-JWT, now RFC 9901, defines how to publish a signed token in which individual claims can be withheld at presentation. SD-JWT VC builds a credential format on it and introduces a type identifier: "This specification defines the new JWT claim vct (for verifiable credential type). Its value MUST be a case-sensitive string serving as an identifier for the type of the SD-JWT VC." It is text-based, familiar to anybody who has worked with web tokens, and much easier to adopt for an existing web service.
And the W3C data model sits alongside both. The Verifiable Credentials Data Model 2.0 became a W3C Recommendation on 15 May 2025 and defines the abstract model — "A verifiable credential is a specific way to express a set of claims made by an issuer, such as a driver's license or an education certificate." It is the vocabulary rather than the wire format, which is why a deployment can be "verifiable credentials" and still require a decision between the two formats above.
The constraint that catches people. The European reference framework states it plainly: "SD-JWT VC-compliant attestations cannot be used in proximity presentations." If you need a credential presented to a device across a counter, with no network, the JSON token route does not serve you and the ISO mdoc does.
What the interoperability profile requires. The high assurance profile settles the choice by requiring support for at least one: "The Wallet and Verifier MUST support at least one of the following Credential Format Profiles defined in Section 6: IETF SD-JWT VC or ISO mdoc."
So the decision. If your use is online only — a web or app journey, over the internet — the JSON token is the lower-friction path. If any part of your use is in person, you need the ISO format, and if you have both, you need both. Choosing one and discovering the other requirement later is a rebuild, not a configuration change, because the cryptographic containers and the presentation protocols differ.
Issuance and presentation: the protocols
The format says what a credential looks like. Two separate specifications say how it gets into a wallet and how it comes out.
Issuance. OpenID for Verifiable Credential Issuance, final as of September 2025, covers the flow from an issuer to a wallet. Two details that shape an implementation:
- Batching. The specification defines a batch size parameter: "REQUIRED. Integer value specifying the maximum array size for the proofs parameter in a Credential Request. It MUST be 2 or greater." Issuing several credentials at once is a first-class operation, and lesson eight explains why an issuer would want to — it is the mechanism behind one of the unlinkability designs.
- A transaction code, optional, "indicating that a Transaction Code is required if present, even if empty" — the out-of-band step that binds an issuance to a person who was present, which matters when the credential is being issued remotely.
And one statement of responsibility worth quoting to anybody who assumes a credential is fire-and-forget: "The Credential Issuer is supposed to be responsible for the lifecycle of its Credentials. This means the Credential Issuer will invalidate Credentials when it deems appropriate, e.g., if it detects fraudulent behavior."
Presentation. OpenID for Verifiable Presentations, final as of July 2025, covers the request from a verifier and the response from a wallet. The requirement that does the security work is the freshness of the challenge: "The Verifier MUST create a fresh, cryptographically random number with sufficient entropy for every Authorization Request, store it with its current session, and pass it in the nonce Authorization Request Parameter to the Wallet."
That is the same mechanism as a web authentication challenge, for the same reason: without it, a captured presentation can be replayed. A verifier that reuses a nonce, or does not bind it to the session, has built a system that accepts a recording.
And a requirement about behaviour rather than cryptography. The same specification: "The Verifier SHOULD ensure that the purpose for collecting the information it is requesting is sufficiently specific and communicated before collection." The holder is being asked to disclose; they are entitled to know what for, before deciding. Lesson nine returns to this as a legal obligation rather than a recommendation.
Why a profile exists at all. Two implementations can each conform to these specifications and fail to interoperate, because the specifications leave choices open. The high assurance interoperability profile closes them — formats, as the previous lesson showed, and also constraints such as: "Each Credential MUST have its own unique, unpredictable status list index, even when multiple Credentials reference the same status list URI."
If you are procuring a wallet or a verifier, conformance to the profile is a more useful contractual requirement than conformance to the underlying specifications, because the profile is what makes two products work together.
In person and over the internet are different problems
A credential presented across a counter and the same credential presented to a website are different engineering exercises with different standards and different attacks.
Two standards, deliberately separated. ISO/IEC 18013-5 covers presentation in person: a reader device and the holder's phone exchanging data directly, typically over a short-range radio link, with no assumption that either is online. ISO/IEC TS 18013-7 adds the other case, augmenting the licence standard with "presentation of a mobile driving licence to a reader over the internet."
The separation is not administrative. In the proximity case the verifier can see the person, the device is physically present, and the channel is local. Online, none of those is true, and everything that was answered by physical presence has to be answered cryptographically.
A status detail worth having right. The 2024 edition of the online standard has been withdrawn and replaced by a 2025 edition. Anybody working from a reference list compiled in 2024 is pointing at a superseded document — and as lesson eleven shows, so is a piece of European legislation.
What proximity gives you, and what it costs. It gives presence: a person standing there, whose face can be compared with a portrait in the credential. The reference framework notes that the ISO standard "was originally developed as a standard for mobile driving licences (mDL) and mDL readers", which is why the model fits a roadside check or a bar.
What it costs is reader deployment. Every verification point needs hardware or a capable phone with the right software, and the offline case means the reader must hold the trust material locally and have a policy for how stale that material may be.
What online gives you, and what it costs. It gives reach: any website or application can be a verifier, with no hardware. What it costs is every assurance that presence used to provide. There is nobody to look at, so holder binding is entirely cryptographic — lesson five. The device may be an emulator. And the request may be made by a party the holder cannot see or assess, which is the whole subject of lesson nine.
One asymmetry to design around. The attacks differ. In person, the threat is a device showing something it should not — a cloned or forged credential, or a borrowed phone. Online, the threat is the verifier as much as the holder: a relying party that requests more than it needs, or a convincing imitation of a legitimate one.
So the controls are not symmetrical either. Proximity needs reader-side validation and a face comparison. Online needs verifier registration, purpose limitation and a fresh challenge bound to the session.
And a practical note for anybody planning both. The reason the format decision in lesson two is a rebuild rather than a configuration is sitting in this lesson: the JSON token format cannot be presented in proximity at all. Decide which journeys you will support before you choose the format, not afterwards.
Trust moves, it does not vanish
The sentence that misleads people about this technology is that it removes the need for trusted intermediaries. It does not. It relocates the trust from a document's physical features to an issuer's signature, and a signature is only worth what the issuer is worth.
The failure mode, stated bluntly. A credential signed by an issuer you should not trust verifies perfectly. The cryptography is correct, the signature checks, the claims are intact, and the whole thing is worthless. Verification answers "was this signed by the holder of key X", and the question you actually have is "should I believe what key X asserts".
So a verifier needs a trust list. A list of the issuers it will accept, and for each, which claims it will accept from them. A driving authority is authoritative about a driving entitlement and not about an employment status. An issuer trusted for one claim should not be trusted for all claims, and a verifier that checks only the signature has conflated the two.
Where the trust list comes from. In the European model it comes from a registration regime: issuers and wallet providers are supervised, and the trust material is published for verifiers to consume. The reference framework is explicit that this is procedural rather than cryptographic: "This trust is primarily rooted in authority and in procedural measures, such as public oversight, published security and operational policies, and audits, rather than in technical measures."
Read that twice if you came to this expecting the technology to replace institutions. The cryptography makes the assertion tamper-evident. Whether the assertion means anything rests on oversight, policy and audit — the same apparatus as before, with better plumbing.
The operational consequences. A trust list is a live dependency with the properties of any other:
- It changes. Issuers are added, suspended and removed, and a verifier caching it for a month will accept a suspended issuer for a month.
- It must be available offline for the proximity case, which means a local copy, a staleness policy, and a decision about what to do when the copy is too old. Accepting anyway and failing closed are both defensible; not having decided is not.
- It is a single point of compromise. A verifier that can be made to accept an attacker's issuer has been fully defeated with no cryptography broken at all.
- It needs a fallback. What a verifier does when the list cannot be reached is a policy question that will be asked of it on the first bad day.
And the parallel worth drawing. This is the certificate authority problem in new clothing. The same questions arise — who is on the list, who decides, how is removal propagated, what happens when a trusted party misbehaves — and the answers are no easier because the credentials are newer. An institution that has managed a certificate trust store already knows most of this work.
Does the credential belong to the person presenting it?
A verified credential is a verified statement about a subject. It is not, on its own, evidence that the person presenting it is that subject. Separating those two is the most common conceptual error in deployments.
Three things that must each be established.
- Issuer authenticity. The signature verifies against a key you trust for these claims. Lesson five.
- Holder binding. The credential is cryptographically bound to the device presenting it, and that device can prove it holds the private key.
- Subject identity. The person operating the device is the subject the credential describes.
Holder binding, mechanically. SD-JWT requires that "When Key Binding is required by the use case, the SD-JWT MUST contain information about the key material controlled by the Holder", and verification uses that: "To verify the KB-JWT, the cnf claim of the SD-JWT MUST be used." The credential carries a reference to a key; the presentation is signed with that key; the verifier checks both.
The European rules place the same obligation on the issuer and the wallet. Providers of person identification data "shall ensure that person identification data that they issue is cryptographically bound to the wallet unit to which it is issued", and wallet providers must ensure wallet units "support proving the possession of private keys corresponding to public keys used in cryptographic bindings". Wallets must also "generate a proof of possession of private keys".
Together these mean a copied credential file is useless: without the private key, the holder binding cannot be demonstrated.
And subject identity, which cryptography cannot settle. The presentation specification names the mechanism and the limitation in one line: "One example of a Credential with biometric Holder Binding is a mobile driving license [ISO.18013-5], which contains a portrait of the Holder."
A portrait in the credential lets a human or a matcher compare the person to the signed image. That works in proximity, where there is a person to look at. Online it requires a biometric comparison of a live capture against the portrait — which is exactly the face matching and liveness detection the eKYC courses cover, reappearing inside the new stack.
So the honest summary. The new model removes the document inspection problem and the data-entry problem. It does not remove the biometric problem. A system that accepts a valid credential from an unlocked phone, with no comparison of any kind, has verified a phone.
Where that is acceptable. For a low-consequence claim — proving an age band to enter a venue — binding to the device plus the device's own unlock may be proportionate, and demanding a face comparison would be over-identification. For opening an account or authorising a payment it is not. The decision belongs to the relying party and it should be written down, which is the record lesson ten covers.
Revocation, and why it is still the hard part
A credential valid at issuance may be cancelled tomorrow. Establishing whether it has been is the oldest unsolved problem in this field, and the new stack has not solved it either — it has made the tension visible.
Why it is hard here specifically. The conventional answer is to ask the issuer. But the issuer learning that this holder is presenting a credential, right now, to this verifier, is exactly the surveillance the architecture is supposed to prevent. Privacy and freshness pull against each other, and every design is a position on that trade.
The W3C answer: a bitstring. The Bitstring Status List recommendation, also a W3C Recommendation as of 15 May 2025, "describes a privacy-preserving, space-efficient, and high-performance mechanism for publishing status information such as suspension or revocation of Verifiable Credentials through use of bitstrings."
The privacy comes from size. "The default status list size is 131,072 entries, equivalent to 16 KB of single bit values", and the reason is stated outright: "This specification uses a minimum list length of 131,072. This size ensures an adequate amount of group privacy in the average case."
The mechanism is herd immunity. A verifier downloads the whole list and reads one bit from it, so the issuer learns that somebody fetched the list and not which credential was checked. Sixteen kilobytes is a small price for that.
The IETF answer, still a draft. The Token Status List specification takes the same approach with a token wrapper, allowing more than one bit per entry so that states beyond valid and revoked can be expressed: "The allowed values for bits are 1, 2, 4, and 8." Retrieval is a plain fetch: "The Status Provider MUST return the Status List Token in response to an HTTP GET request to the URI provided in the Referenced Token."
It is in the RFC Editor's queue and does not yet have a number. Lesson eleven is about what that should mean for your plans.
The residual leak, which the drafts admit. "Two or more colluding parties (e.g Relying Parties and or the Status Issuer) may link two transactions involving the same Referenced Token by comparing the status claims of received Referenced Tokens." The index is itself an identifier. Two verifiers who compare notes can tell they saw the same credential, which is why the interoperability profile insists that each credential have its own unique, unpredictable index.
The alternative that avoids the problem entirely. Issue credentials that expire before revocation would matter. The European discussion paper puts it as a design option: "Only issue short-lived attestations having a validity period of 24 hours or less, such that revocation will never be necessary."
That trades a hard distributed-systems problem for a different one — the wallet must be able to refresh daily, which requires the issuer to be available and the holder to be online. For a credential used in a connected journey that is a good trade. For one that must work in a rural area with no signal it is not, and that is the choice to make deliberately rather than by default.
Unlinkability, and the privacy you have to design in
The claim that this technology is privacy-preserving is true of what it makes possible and not of what it does by default. Several of the properties people assume are present have to be chosen, and one of the most important is not yet widely deployable.
The property, named. The European discussion paper defines the risk: "Relying Party linkability means that a link can be established between multiple presentations of attestations by the same User to one or more Relying Party." Two verifiers, or one verifier across two visits, determining that they saw the same person.
What selective disclosure does and does not give you. It reduces what you hand over — proving an age band without the date of birth, residence without the address. The W3C model states the practice: "One accepted best practice for preventing such a violation is for verifiers to limit the information requested and received, to the absolute minimum necessary for a particular transaction."
But a signature is itself an identifier. Present the same signed credential twice and the two presentations carry the same bytes, whatever claims you withheld. Selective disclosure is data minimisation; it is not unlinkability.
What SD-JWT does about the salts. Each withheld claim is salted, and the specification requires "The salt value MUST be unique for each claim that is to be selectively disclosed. The Issuer MUST NOT reveal the salt value to any party other than the Holder." That prevents a verifier learning the content of claims it was not shown. It does not prevent two verifiers recognising the same credential.
The specification is candid about the goal: "Unlinkability is a property whereby adversaries are prevented from correlating credential presentations of the same user beyond the user's consent."
Three approaches, in ascending order of how deployable they are today.
- Pairwise identifiers, which is the oldest answer and applies where a decentralised identifier is used: "DID controllers can mitigate this privacy risk by using pairwise DIDs that are unique to each relationship; in effect, each DID acts as a pseudonym." A different identifier per relying party, so there is nothing common to join on.
- Present each credential once. The blunt and effective method: "In this approach, the Attestation Provider requires the Wallet Unit to present each attestation only once." The wallet holds a batch of single-use credentials and spends one per presentation — which is why the issuance protocol in lesson three makes batching a first-class operation.
- Blinded signatures, which solve it properly. The W3C model: "Blinded signatures allow for unlinkable disclosure, which remove a common source of holder correlation during multiple presentations to one or more verifiers." This is the cryptographically clean answer and it is not yet available in the mainstream formats, which is the honest state of the art.
And a wallet-level requirement worth knowing. The interoperability profile addresses the wallet's own attestation: "Wallet Attestations MUST NOT be reused across different Issuers. They MUST NOT introduce a unique identifier specific to a single Wallet instance." Without that, the wallet itself becomes the correlating identifier and everything above is wasted.
The verifier's conclusion. Ask for less, do not retain what you did not need, and do not assume the architecture has made you unable to track people. It has made it possible for you not to.
What assurance can you actually claim?
A relying party that accepts a wallet credential has to be able to say what level of identity assurance it obtained. Two frameworks answer in different vocabularies, and both matter if you operate across jurisdictions.
The NIST ladder, and where remote proofing sits. SP 800-63A-4 describes IAL2 as requiring "collecting additional evidence and more rigorous processes for validating evidence and verifying identities, including enhanced processes to confirm that the applicant is the rightful owner of the presented evidence." IAL3 goes further, adding "the requirements for a trained CSP representative (i.e., proofing agent) to interact directly with the applicant as part of an on-site attended identity proofing session and the collection of at least one biometric characteristic."
Note that the top level is defined around attended proofing. A wallet credential presented over the internet does not reach it by being cryptographically excellent, because the level is about the process by which identity was established, not about the strength of the container.
And where a mobile licence counts as evidence. This is the most directly useful sentence in the whole framework for anyone building this. SP 800-63A-4's table of superior evidence examples lists: "Mobile Driver's License (MDL) State issuance processes, AAMVA guidance, or Real ID Act • Validation of mobile security object • Revocation check if available".
Three things are being asked for there, and only the first is about the credential. The issuance process behind it has to meet a standard. The security object has to be validated — the cryptographic check. And a revocation check is expected where one is available, which is lesson seven reappearing as an assurance requirement rather than an engineering preference.
An institution that verifies the signature and skips the status check has not met the condition attached to treating the credential as superior evidence.
The European ladder. The eIDAS framework uses assurance levels low, substantial and high, and the wallet rules require the high level: wallet units must achieve "assurance level high, as set out in Implementing Regulation (EU) 2015/1502". The level attaches to the wallet solution rather than to each presentation, which is a different model from the NIST one — a supervised wallet at a stated level, rather than a per-transaction assessment.
How to reconcile them in a file. Do not map one to the other and claim equivalence; the frameworks measure different things. Record, per journey: which credential type was presented, which issuer signed it, that the signature and the status were checked and when, how holder binding was established, whether a biometric comparison was performed, and which framework's level you are asserting and on what basis.
And the discipline the eKYC courses already teach, unchanged. An assurance level with no record behind it is a label. The new stack changes what the record contains — a signature chain and a status check rather than image quality scores and match thresholds — and changes nothing about the need to keep it, or about the fact that someone will read it years later.
The verifier's obligations, and what goes in the file
In the European model the relying party is regulated, not merely served. That is a change from the document-checking world, where nobody licensed the person looking at your passport.
You may not ask for more than you registered for. The regulation is short and absolute: "Relying parties shall not request users to provide any data other than that indicated pursuant to paragraph 2, point (c)." A relying party declares what it needs, and asking for more is a breach rather than a negotiation with the holder.
This inverts a long-standing habit. In an onboarding flow it has always been cheaper to collect everything and decide later what to use. Here the request itself is bounded by what you declared, and the wallet is positioned to refuse what you did not.
And the purpose has to be communicated before collection. The presentation specification: "The Verifier SHOULD ensure that the purpose for collecting the information it is requesting is sufficiently specific and communicated before collection." A holder deciding whether to disclose is entitled to know what for, at the moment of deciding, not in a privacy notice they will never open.
Over-identification is the live risk, and it is yours. The technology makes it trivially easy to request a full set of attributes, because the request is a parameter rather than a form a customer has to fill in. The constraint is therefore legal and procedural rather than practical, which means it depends on somebody inside the organisation deciding what the minimum is and holding the line when a product manager asks for date of birth "for analytics".
What the file has to contain, and this is the part that outlives everyone involved:
- The request you made: which attributes, under which declared purpose, with the registration that permitted it.
- What you received, and nothing you did not need. A log that records the full credential when you asked for one attribute has defeated the minimisation you just implemented.
- The checks performed: signature verified against which trust list, as of which version; status checked by which method, at what time; holder binding demonstrated how.
- Any biometric comparison: that it happened, what it concluded, and the threshold in force — recorded as a decision, not as an image you now have to protect.
- The version of everything: the credential format, the protocol profile, the trust list, and your own rules. A decision is only explicable against the rules that applied on the day.
- The outcome and who or what decided it, including whether a person reviewed it.
And the retention question, which is specific here. You are now holding less personal data than a scan-and-store pipeline would have given you, which is the point. Do not undo it in the logs. The temptation to keep the whole presentation "in case of a dispute" converts a privacy-preserving architecture into a conventional data store with extra cryptography, and the eKYC and data protection courses both reach the same conclusion from their own directions.
Final, draft, and where the law disagrees with the standards
This field moves faster than its documentation, and two kinds of mismatch will cost you if you do not check for them.
First: not everything load-bearing is finished. Two specifications this course has quoted are drafts.
SD-JWT VC — the credential format, one of the two the interoperability profile requires — is an Internet-Draft. At the time of writing it is at revision 26 and in late review, which is encouraging and is not the same as final. Token Status List, from lesson seven, is in the RFC Editor's queue and has no RFC number.
Both are mature, both are being implemented, and both can still change. The professional position is not to avoid them — you cannot build this without them — but to know which parts of your system rest on a draft, so that when a revision lands you know what to re-test. A dependency list that does not distinguish a Recommendation from a draft-26 cannot answer that question.
By contrast, RFC 9901 for SD-JWT itself, the Verifiable Credentials Data Model 2.0, Decentralized Identifiers 1.0, Bitstring Status List, and the OpenID issuance and presentation specifications are all final.
Second: the legislation names superseded standards. This is the sharper problem, because it is not a judgement call.
The European implementing regulation on person identification data requires two formats: "the format specified in ISO/IEC 18013-5:2021" and "'Verifiable Credentials Data Model 1.1.', W3C Recommendation, 3 March 2022." Version 1.1 — while 2.0 has since become the W3C Recommendation.
And the implementing regulation on protocols lists in its annex: "ISO/IEC 18013-5:2021 — ISO/IEC TS 18013-7:2024". The 2024 edition of that technical specification has been withdrawn by ISO and replaced by a 2025 edition.
What to do about it, practically. You cannot resolve a conflict between a regulation and a standards body by choosing the one you prefer. What you can do:
- Implement what the law names, where the law applies to you, because that is the compliance obligation.
- Record that you know it is superseded, and why you chose as you did. A file that shows awareness is a different conversation with a supervisor from one that shows ignorance.
- Watch for the amendment, because legislation of this kind is updated by implementing act, and the reference will be corrected.
- Isolate the version-specific code so that the correction is a configuration change rather than a project.
And the general lesson, which applies beyond this field. A reference list is a dated artefact. Two of the documents this course cites were withdrawn and replaced while the specifications around them stayed still, and one piece of binding legislation now points at a withdrawn document. Check the status of a standard at the moment you rely on it, not at the moment you first read it.
Losing the phone, the deadline, and what to do now
Three practical matters that pilots discover late.
The phone will be lost. The whole model puts the credential on a device, so the device going missing is the ordinary case rather than an edge one, and the rules address it directly. Wallet providers "should enable wallet users to request the revocation of their wallet unit" where users "lost their wallet units or no longer have control over it". The reference framework spells out the consequence: "The User needs a User account at the Wallet Provider to ensure that they can request the revocation of their Wallet Unit in case of theft or loss."
Read that carefully, because it reintroduces something the architecture was supposed to remove: an account, with a provider, protected by credentials of its own. That account is now part of the security of every credential in the wallet, and the recovery lesson from the authentication course applies to it in full.
Recovery is also contemplated — "making it possible to recover lost wallet units or to transfer information from one wallet provider to another, where appropriate" — and portability between providers is a resilience property worth asking a vendor about rather than assuming.
The deadline is a computation, not a date. The regulation requires that "each Member State shall provide at least one European Digital Identity Wallet within 24 months of the date of entry into force of the implementing acts referred to in paragraph 23 of this Article and in Article 5c(6)."
Twenty-four months from the entry into force of the implementing acts. Those were published on 4 December 2024 and enter into force on the twentieth day after publication, which puts the obligation around December 2026. No document in this course prints that as a date — it is arithmetic, and it is presented here as arithmetic. If you need a date for a plan, take it from the Commission's own statement rather than from this calculation.
And the other hemisphere. In North America the route is the AAMVA mobile driving licence implementation guidelines, now at version 1.6, whose published summary notes revisions to the over-the-internet appendix, versioning moved to its own chapter, and "updating mDL revocation to align with issuing authorities' processes". Its stated aim is "Technical interoperability between different issuing authorities' mDL programs, enabling an issuing authority to read an mDL issued by any other issuing authority." The guidelines document itself is not freely retrievable, so this course cites the announcements and not the clauses.
What to do now, in order. For an institution that will be a verifier rather than an issuer:
- Decide which journeys need proximity and which are online, because that decides the format and the format is hard to change.
- Require profile conformance in procurement, not specification conformance.
- Build the trust list as a managed dependency, with a staleness policy and a documented fallback.
- Decide your subject-identity rule per journey — when device binding is enough and when a face comparison is required — and write it down.
- Keep the minimisation in the logs, not only in the request.
- And keep the old route working. Adoption is uneven and will be for years, so chips, faces and credentials will be verified side by side. Everything the eKYC courses teach stays load-bearing.
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.
Your answers
Your certificate of completion
Enter your name as you want it to appear, then save the certificate as a PDF. In the print window, choose Save as PDF. A certificate is issued once per completion of the course.
Rate this course
How would you rate this course? Your rating is anonymous — just a number, with no name and no account.
Bring this to your team
Your colleagues can take this course too — it is free, needs no sign-up, and nothing they enter leaves their browser. If your institution wants the same ground covered as structured training for a team, with an instructor and your own cases, we run courses in a classroom or online for groups of ten or more.
Talk to us about training your team → See the paid courses →
Saolix does not record who takes this course, so it cannot verify these certificates. The certificate confirms completion of a free self-paced course and is not an accredited qualification.
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.
- ISO/IEC — ISO/IEC 18013-5:2021: Mobile driving licence (mDL) application (catalogue entry and abstract), ISO/IEC TS 18013-7:2025: mDL add-on functions, online presentation (catalogue entry and abstract)
- W3C — Verifiable Credentials Data Model v2.0 (W3C Recommendation, 15 May 2025)
- IETF — RFC 9901: Selective Disclosure for JSON Web Tokens (SD-JWT, November 2025)
- NIST — SP 800-63A-4: Digital Identity Guidelines, Identity Proofing and Enrollment (July 2025)
Sources
The official documents this course relies on. Laws and guidance change, so check the current version.
- ISO/IEC 18013-5:2021: Mobile driving licence (mDL) application (catalogue entry and abstract) · International Organization for Standardization
- ISO/IEC TS 18013-7:2025: mDL add-on functions, online presentation (catalogue entry and abstract) · International Organization for Standardization
- Verifiable Credentials Data Model v2.0 (W3C Recommendation, 15 May 2025) · World Wide Web Consortium (W3C)
- Decentralized Identifiers (DIDs) v1.0 (W3C Recommendation, 19 July 2022) · W3C
- Bitstring Status List v1.0 (W3C Recommendation, 15 May 2025) · W3C
- RFC 9901: Selective Disclosure for JSON Web Tokens (SD-JWT, November 2025) · Internet Engineering Task Force (IETF)
- SD-JWT-based Verifiable Digital Credentials (SD-JWT VC) - Internet-Draft, not yet an RFC · IETF
- Token Status List - Internet-Draft, in the RFC Editor queue · IETF
- OpenID for Verifiable Presentations 1.0 (final, 9 July 2025) · OpenID Foundation
- OpenID for Verifiable Credential Issuance 1.0 (final, 16 September 2025) · OpenID Foundation
- OpenID4VC High Assurance Interoperability Profile 1.0 (final, 24 December 2025) · OpenID Foundation
- Regulation (EU) 2024/1183: the European Digital Identity Framework, amending eIDAS · Official Journal of the European Union, via EUR-Lex
- Commission Implementing Regulation (EU) 2024/2977: person identification data and electronic attestations of attributes · European Commission, via EUR-Lex
- Commission Implementing Regulation (EU) 2024/2979: integrity and core functionalities of wallets · European Commission, via EUR-Lex
- Commission Implementing Regulation (EU) 2024/2982: protocols and interfaces · European Commission, via EUR-Lex
- EUDI Wallet Architecture and Reference Framework, chapters 5 and 6 · European Commission
- EUDI Wallet ARF Discussion Topic A: privacy risks and mitigations (discussion paper, not normative) · European Commission
- SP 800-63A-4: Digital Identity Guidelines, Identity Proofing and Enrollment (July 2025) · US National Institute of Standards and Technology
- Now Available: AAMVA's Mobile Driver License (mDL) Implementation Guidelines v1.6 (30 June 2026) · American Association of Motor Vehicle Administrators



