Two flaws in Credo, the open-source framework hosted by the OpenWallet Foundation that wallet and verifier apps use to check digital credentials, could let data that the named issuer never signed be treated as verified: in one case data nobody had signed, in the other data signed by someone other than the issuer it claimed. Both were reported privately, fixed by the Credo maintainers in Credo 0.7.5 on 18 November 2026, and published as two security advisories that list this researcher (GitHub: otoinobuo) as the reporter.
In plain terms
A digital credential is the electronic form of an ID card, licence or certificate, held in a phone wallet and checked by an app. The check is what stops someone from simply typing in the details they want.
- A note slipped in next to a real ID was read as the ID. A person could hand over one genuine credential together with a made-up block of details, for example a different name or an "admin" role. Credo checked the genuine credential, then gave the app the made-up block as if it were the checked one. An app that took its details from that result could end up acting on details no issuer had signed.
- A signature was accepted without checking whose it was. For one kind of public record (schemas and credential definitions published under a
did:webvhidentifier), Credo accepted a record in an issuer's name as long as it carried a valid signature from any did:webvh identity, not necessarily the issuer's. The acceptance was shown in Credo's code; the maintainers call it theoretical because a real attack also needs control of where that issuer's records are served from, and records already trusted stay protected by their content hashes.
Reporting privately let the maintainers fix both before the details were public. Anyone running an app built on Credo can remove both problems by updating to 0.7.5.
The problems
1. OpenID4VP with Presentation Exchange: an unsigned object returned as the verified credential (GHSA-h288-6jfw-x36p)
When Credo verified an OpenID4VP response that used DIF Presentation Exchange, it returned a credential for each requested item. For W3C presentations it found that credential by following a path written by the holder, and the path could point outside the presentation's list of credentials. The request's requirements were checked against a different credential. A holder with one valid credential could therefore add an object with any claims and any issuer, point the path at it, and have Credo return it as the credential for that item. The object carries no issuer signature and its issuer is not checked against the trusted issuers.
Affected: Credo 0.6.0 to 0.7.4, in applications that read descriptors[].credential from Presentation Exchange. Requests that use DCQL, and DIDComm proof exchange, are not affected.
2. did:webvh attested resources: the signer was not tied to the named issuer (GHSA-xwfx-6wv7-j7p7)
When Credo fetched a schema, credential definition or revocation record published as a did:webvh attested resource, it verified the signature against the signer's own identifier but never checked that the signer was the issuer the record names. The issuer check compared two fields that the record itself supplies. A party with their own valid did:webvh key could sign a record in another issuer's name, and Credo accepted it.
Affected (per the advisory): @credo-ts/webvh and @credo-ts/core before 0.7.5. The advisory rates it Moderate (CVSS 5.9) because the attacker must also be able to serve bytes at the victim issuer's URL, for example by compromising the host or the network path; on hosting shared by several issuers, one compromised host could affect every issuer it serves.
How it was shown
Both were reproduced against Credo's own code rather than argued from reading. For the first, a presentation signed with a real Ed25519 key passed Credo's own presentation verifier and Presentation Exchange check, and Credo's lookup for the requested item returned the added, unsigned object carrying role: admin (the HTTP transport of OpenID4VP was not exercised). For the second, Credo's own proof verifier and registry accepted a schema in the victim issuer's name signed with the attacker's key (only DID resolution, the wallet and the network fetch were stubbed), while altered data and unauthorised keys were still rejected. Both were reported through GitHub's private vulnerability reporting; for the did:webvh report the full reproduction was shared with the maintainers in the private advisory.
Response
The maintainers accepted both reports and wrote the fixes. The issuer check was fixed in pull request #3237 the same day, about eight hours after the report (12 Nov 2026). The path lookup was fixed in pull request #3267 (on main, backported to the 0.7.x line), which accepts only paths that point at one of the presentation's own credentials and checks the request against that credential. Both fixes shipped in Credo 0.7.5 on 18 November 2026, and the two advisories were published the same day, rated Moderate (CWE-345), each listing GitHub user otoinobuo as the reporter. No CVE had been assigned as of 18 Nov 2026.
Links
Cite as
Nobuo Otoi (ba-cybersecurity.com). Credo: two credential-verification flaws (unsigned data returned as a verified credential; issuer impersonation for did:webvh resources), report and fix. 2026. https://ba-cybersecurity.com/research/credo-ts-credential-verification/.
@misc{otoi2026credo,
author = {Otoi, Nobuo},
title = {Credo: two credential-verification flaws (unsigned data returned as a verified credential; issuer impersonation for did:webvh resources), report and fix},
year = {2026},
howpublished = {\url{https://ba-cybersecurity.com/research/credo-ts-credential-verification/}},
note = {Advisories: GHSA-h288-6jfw-x36p, GHSA-xwfx-6wv7-j7p7 - Fixed in Credo 0.7.5}
}
Reported 12 and 15 Nov 2026 - Fixed and released 18 Nov 2026 (0.7.5) · This page: updated 18 Nov 2026 · Permanent address: https://ba-cybersecurity.com/research/credo-ts-credential-verification/