Two privacy properties are confused more often than any other pair in the credentials field, including in policy documents: selective disclosure and unlinkability. They are different properties, delivered by different mechanisms, and since 2024 they carry different legal weight in Europe. This page separates them, shows how each of the two stacks handles them, and lists what a data protection officer should check. It draws on the VC guide v1.1 (EUNIS SIG, July 2026), Verifiable Credentials in Higher Education: W3C-VC and SD-JWT VC, published by the EUNIS Special Interest Group – Mobility & Digital Credentials (EUNIS SIG).
Two properties, not one
Selective disclosure lets the holder show a subset of the attributes in a credential and keep the rest back: a graduate can prove she has a Master’s degree while withholding her date of birth. Both stacks offer it.
Unlinkability means that two presentations of the same credential cannot be correlated with each other: not by one verifier who sees the holder twice, and not by several verifiers comparing notes. It is a stronger property. A credential can offer perfect selective disclosure and still be perfectly linkable: if some element of every presentation is byte-identical, that element works as a persistent pseudonym for the holder, whatever attributes were hidden.
Put simply: selective disclosure limits what a verifier learns in one presentation; unlinkability limits what anyone can learn by joining presentations together.
How each stack behaves
SD-JWT: salted hashes under one stable signature
In an SD-JWT, each disclosable claim is replaced in the signed payload by a salted hash, and the issuer signs the whole. The holder reveals only the claims a verifier needs. The mechanism is elegant and does exactly what it promises. But the issuer’s signature and the salted hashes are the same in every presentation. Any verifier who sees the credential twice, or any two verifiers who compare records, can correlate the presentations.
The usual mitigation is batch issuance: the issuer mints many single-use copies, each with its own signature and salts. It helps, with documented limits. Someone must anticipate how many presentations the credential will need over its life; the wallet must manage the stock; the holder keeps a lifelong channel open to the issuer to replenish it; and it offers no protection if a verifier and the issuer compare notes, because the issuer knows every copy it minted.
W3C-VC with bbs-2023: a fresh proof every time
In the W3C-VC stack the proof is a separate layer, and one of the Data Integrity cryptosuites, bbs-2023, changes the picture. The issuer signs once; the holder then derives a mathematically fresh proof for each presentation, revealing only the chosen attributes. Two derived proofs of the same credential are cryptographically unrelated, even if verifiers collude and even with the issuer’s cooperation. At the level of the proof, unlinkability here is a property of the mathematics, not a promise in a policy.
Honest caveats belong next to that claim. BBS unlinkability holds per derived proof. It does not anonymise attribute values that are themselves identifying. If the holder discloses her full name, a national identifier or a rare combination of attributes, verifiers can link her through those values regardless of the cryptography. BBS makes each proof fresh, but metadata can still correlate: a status-list index or a unique credential identifier disclosed in every presentation is a stable handle unless the design hides it (for example with large, cached status lists). bbs-2023 removes the correlation handle the proof itself would otherwise create; it does not remove the ones the holder chooses to reveal or the ones the credential’s design leaves in place. Wallet support for bbs-2023 is also still maturing, a point covered under Open problems.
Side by side
| Property | SD-JWT | W3C-VC with bbs-2023 |
|---|---|---|
| Selective disclosure | Yes, via salted-hash disclosures | Yes, via derived proofs |
| Unlinkability across verifiers | No: issuer signature and salted hashes are stable across presentations | Yes: each presentation carries a freshly derived proof |
| Protection against verifier–issuer collusion | No, even with batch issuance | Yes for the proof itself; status-list indices and identifiers must also be designed not to correlate |
| Main mitigation | Batch issuance of single-use copies | Not needed for the proof; data minimisation still needed for identifying values |
| What the institution must operate | Batch sizes, replenishment channel and stock management for the credential’s lifetime | Status-list operation and design choices that avoid correlating metadata |
| Where the guarantee lives | In procedures and operations | In the cryptography of the format |
| Maturity today | Mature, widely supported in wallets | Wallet support still maturing |
What the law asks, in plain words
The European Digital Identity Wallet is governed by eIDAS 2.0. Article 5a(16) of Regulation (EU) 2024/1183 sets two requirements for the wallet’s technical framework. Point (a): after issuance, neither the attestation provider nor any other party may obtain data that lets transactions or user behaviour be tracked, linked or correlated, unless the user explicitly authorises it. Point (b): the framework must enable privacy-preserving techniques that ensure unlinkability where the attestation does not require identifying the user. The target is unlinkability, not merely selective disclosure; and where a presentation must identify the holder, as many degree checks do, point (a) carries the weight. Treating the two properties as synonyms is not a small technical slip; it is a misreading of what the regulation names.
Article 25 GDPR (data protection by design and by default) asks controllers to build data protection principles, such as data minimisation, into the processing itself, not to bolt them on afterwards through organisational measures. Article 5(2) GDPR (accountability) asks controllers to be able to demonstrate compliance. Read together, they reward architectures where privacy properties can be shown in the design, and make life harder for architectures where they can only be asserted in procedures.
The practical difference shows up in a supervisory inspection. One stack lets the DPO point to a format property; the other requires the DPO to defend a set of procedures, such as batch policies and replenishment rules, and evidence that they were followed. Both can be compliant. They are not equally easy to defend. This page explains the reasoning; it is not legal advice, and each institution should take its own legal view.
Aina’s career, presentation by presentation
Take the site’s worked example, a hypothetical scenario: Aina’s Master’s in Computing (EQF 7), issued by URV in 2026. In the scenario, over the following decades she presents it to employers, professional bodies, a foreign university for doctoral admission and a licensing authority.
On SD-JWT she discloses only what each verifier needs, and that works. But the same signature and hashes travel in each presentation, so her degree becomes, in effect, a persistent identifier across her working life. Batch issuance blunts this, at the cost of predicting how many presentations a thirty-year career contains and keeping a channel to the university open for life. On W3C-VC with bbs-2023, every presentation carries an unrelated proof; provided the status-list index and credential identifier are not disclosed as stable handles, nothing further is asked of Aina, of the university after issuance, or of anyone’s good faith. The credential lifecycle page follows this and the other milestones in full.
What a DPO should check
The question the DPO owns is simple: is unlinkability a property of our format, or a promise in our procedures?
- Read the project documentation: if “selective disclosure” and “unlinkability” are used interchangeably, record that as a finding. Correcting vocabulary now is cheaper than correcting architecture later.
- For each credential type, ask which element of a presentation stays constant across verifiers (signature, hashes, identifiers, status index).
- Check correlating metadata on both stacks: is the status-list index or a unique credential identifier (credential id) disclosed in every presentation, and are status lists large and cached so that checking them reveals nothing to the issuer?
- If batch issuance is the mitigation, ask how batch sizes are set, how replenishment works over the credential’s lifetime, and what happens when the stock runs out.
- Ask whether the design protects against a verifier and the issuer comparing notes, not only verifiers among themselves.
- Check which attributes are disclosed by default: even with bbs-2023, identifying values defeat unlinkability.
- Ask what evidence you could show a supervisory authority: an architectural property, or records that procedures were followed.
Which stack to use for which credential is a separate decision, weighed against lifespan, audience and regulatory exposure; see Choosing a stack, which also sets out the EUNIS SIG’s position for formal, regulated credentials.
Key points
- Selective disclosure limits what one presentation reveals; unlinkability prevents presentations from being joined. They are not the same property.
- Both stacks offer selective disclosure. SD-JWT presentations are linkable through a stable signature and salted hashes; batch issuance mitigates but does not resolve this.
- W3C-VC with bbs-2023 gives unlinkability per derived proof, but cannot hide attribute values that identify the holder, and status-list indices or credential identifiers can still correlate unless the design hides them.
- Article 5a(16) of Regulation (EU) 2024/1183 prohibits tracking and correlation of users after issuance and requires the wallet framework to enable unlinkability where an attestation does not require identifying the user; GDPR Articles 25 and 5(2) favour privacy properties that are built in and demonstrable.
