GHSA-H548-CR7V-4V97
Presentation holder-binding / nonce replay protection can be bypassed (SD-JWT VC: KB-JWT optional · mDOC: DeviceAuth never verified)
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:N
Fuente: GitHub
Ciclo de vida de la vulnerabilidad
| Fecha | Fase | Evento | Detalle |
|---|---|---|---|
| Divulgación | Publicada | GHSA-H548-CR7V-4V97 divulgada · CVSS 8.1 (GitHub) | |
| Divulgación | Modificada | Última actualización del registro | |
| Monitoreo | Detectada | Ingresó al índice | |
| Monitoreo | Verificada | Última verificación contra las fuentes |
Summary
`verifyPresentation` / `verifyAuthorizationResponse` return `valid: true` for
presentations that carry **no holder proof-of-possession and are not bound to
the verifier's `nonce`**. Because the issuer-signed material in a presentation
(the SD-JWT issuer JWT + disclosures, or the mDOC `issuerSigned` structure) is
disclosed to *every* verifier and is not secret, an attacker who observes one
presentation — e.g. a malicious or compromised relying party, a logged VP
token, or an intercepted response — can replay it to any other verifier
using this library and be accepted as the legitimate holder, **without ever
possessing the holder's private key and without satisfying a fresh nonce**.
This directly contradicts the README's stated guarantees:
> `nonce` … Yes … Challenge nonce for replay protection
> SD-JWT VC — full cryptographic verification (issuer JWT signature via
> `x5c`, transitive disclosure-hash check, **key binding JWT signature +
> `sd_hash`, nonce check**).
Two independent code paths are affected.
---
Finding A — SD-JWT VC: key binding (KB-JWT) is only checked *if present*
`src/parsers/sd-jwt.parser.ts`, "Step 8":
// Step 8: Verify key binding JWT (if present)
if (parts.kbJwt) {
...
// Validate nonce — must be present and match when KB-JWT is used
if (kbPayload['nonce'] !== options.nonce) {
return invalidResult('Key binding JWT nonce does not match');
}
// Verify sd_hash ...
}Holder proof-of-possession (KB-JWT signature over the holder key in `cnf`),
the `nonce` check, the `sd_hash` binding, and the `audience` check **all live
inside `if (parts.kbJwt)`**. There is no requirement that a KB-JWT be present —
even when the issuer JWT carries a `cnf` claim (i.e. the credential is
explicitly holder-bound). An attacker simply presents `issuerJwt~disc1~disc2~`
with the trailing KB-JWT removed:
- `parts.kbJwt` is `undefined` → the entire block is skipped.
- The issuer signature and disclosure hashes still validate (they are genuine).
- `options.nonce` is never consulted, so the presentation replays across
- Result: `valid: true` with the victim's claims.
### Impact
Complete bypass of holder binding and replay protection for SD-JWT VC. Any
party that has seen a victim's SD-JWT presentation can impersonate the holder to
any verifier built on this library.
---
Finding B — mDOC / ISO 18013-5: DeviceAuth is never verified
`src/parsers/mdoc.parser.ts` verifies `issuerSigned.issuerAuth` (the issuer
COSE_Sign1 over the MSO), MSO validity, and IssuerSignedItem digests — then
returns `valid: true`. It never reads `deviceSigned` / DeviceAuth, never
extracts `deviceKeyInfo.deviceKey` from the MSO, and never binds to a
SessionTranscript / OpenID4VP `nonce`. A grep of `src/` for
`deviceSigned|deviceAuth|sessionTranscript|deviceKey` returns nothing.
ISO 18013-5 §9.1.3 requires the mdoc reader to verify mdoc authentication:
the holder device proves possession of the private key committed in the MSO by
signing (DeviceSignature) or MACing (DeviceMac) the SessionTranscript, which
binds the presentation to the current session/nonce. Omitting this makes the
`issuerSigned` blob a bearer token: anyone who obtains it can replay it.
`options.nonce` (documented as required for replay protection) is accepted but
silently ignored on the mDOC path.
### Impact
Complete absence of holder binding / proof-of-possession for mDOC. A captured
`issuerSigned` structure is replayable to any verifier as `valid: true`.
> Note: the library's own happy-path test fixture (`tests/fixtures/mdoc-helpers.ts`)
> builds `documents: [{ issuerSigned }]` with no `deviceSigned` member, and
> the parser tests assert `valid === true` — confirming device authentication is
> entirely unverified.
---
Proof of Concept
`tests/poc-holder-binding-bypass.spec.ts` (attached) builds a genuine
holder-bound SD-JWT VC, strips the KB-JWT, and shows a replay to a *fresh*
nonce returns `valid: true`; and builds an mDOC with no DeviceAuth and shows it
returns `valid: true`. Both use the library's own test fixtures.
Confirmed on 0.8.1 (vitest 4.1.5) — both assertions pass, i.e. the library
accepts both replayed/holder-unbound presentations:
$ npx vitest run tests/poc-holder-binding-bypass.spec.ts
Test Files 1 passed (1)
Tests 2 passed (2)Each test asserts the *vulnerable* outcome (`expect(replay.valid).toBe(true)` /
`expect(result.valid).toBe(true)`), so a passing run demonstrates the defect.
---
Suggested remediation
- SD-JWT VC: when the issuer JWT contains a `cnf` claim (or when the caller
- mDOC: implement ISO 18013-5 mdoc authentication — parse `deviceSigned`,
- Until fixed, the API should not return `valid: true` for presentations it
## Impact
Prior to 0.9.0, @openeudi/openid4vp did not enforce holder binding, so a verifier could accept a presentation the presenting party does not control — including a captured/replayed one — as if it were a valid proof of possession. This affects both credential formats:
- SD-JWT VC: Key-Binding-JWT verification (signature, nonce, sd_hash, audience) was gated on the KB-JWT being present. An attacker could strip the KB-JWT from a captured credential and have it accepted; holder-bound credentials (issuer cnf) were accepted with no proof of possession.
- mDOC (ISO 18013-5): only the issuer signature (issuerAuth/MSO) was verified. deviceSigned / DeviceAuth / the SessionTranscript / the device key were never checked, so a captured issuerSigned payload was a replayable bearer token.
- The result is an authentication/holder-binding bypass: a relying party could be convinced a subject presented a credential when they did not, or accept a presentation replayed from another session.
## Patches
Fixed in 0.9.0 (breaking — holder binding is now fail-closed for both formats):
- SD-JWT VC: a KB-JWT is mandatory whenever the issuer JWT carries cnf (or requireKeyBinding is set); its signature, a non-empty nonce, sd_hash, and audience are enforced.
- mDOC: full ISO 18013-5 §9.1.3 device authentication — the DeviceSignature over DeviceAuthentication/SessionTranscript is verified against the MSO-committed device key; DeviceMac is rejected; missing deviceSigned/transcript/device-key fail closed.
- Adds OID4VP 1.0-Final OpenID4VPHandover SessionTranscript support. The fix is validated both by the OIDF conformance suite (as an independent ISO 18013-5 wallet) and by the reporter's original proof-of-concept.
## Workarounds
None. Upgrade to 0.9.0 — the missing checks cannot be enabled by configuration in earlier versions.
## Credit
Reported by Emir Çoban (@EmirCobanOfficial), who also independently verified the fix. Thank you.
Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.
Desglose de métricas
Rangos afectados
Estado del parche por paquete
| Paquete / producto | Estado | Corregida en | Desde divulgación | Última versión |
|---|---|---|---|---|
npm@openeudi/openid4vp |
Sin parche | — | — | — |
Arsenal público
Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.
https://github.com/openeudi/openid4vp/security/advisories/GHSA-h548-cr7v-4v97Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.
