...

Once the Google doc version of the specification is sufficiently complete, the specification will move here for final review.

DISCUSSION AREA

This area is for posting and discussing topics relevant to this Task Force.

EU TRAIN Project

TRAIN stands for "TRust mAnagement INfrastructure". It is a subproject run by Fraunhofer-Gesellschaft within the EU eSSIF-Labs Project. This quote is from a recent post to the W3C Credentials Community Group (CCG) mailing list by David Chadwick:

In the TRAIN project we have devised an alternative strategy for trusting VC issuers, based on the existing concept of Trust Lists as standardised by ETSI. It works like this.

Every VC that is issued by any issuer, contains a Terms of Use containing the trust scheme(s) that the issuer is a member of (specified as DNS names). These can be true or false statements, it does not matter.

Any DNS owner can create their own trust scheme and decide which VC issuers are members of it and therefore are trusted to issue VCs of a certain type (with a certain schema). The DNS owner adds a URL RR to their DNS entry containing a TRAIN formatted URL.

The TRAIN open source code will read this RR, dereference the URL and expect to find an ETSI trust list published at this https URL. It will then check if the VC Issuer is listed in this DNS named trust list, and if so will tell the Verifier that the issuer is a trusted member of this trust scheme operated by this "DNS name". In this way it does not matter whether the issuer was telling the truth or not. The TRAIN API and the DNS Owner tells the verifier the truth.

Each entry in the trust list has a Service Type Identifier. This is a URL and the web page pointed to should contain the JSON schema (and @context) for the VCs that are issued for this Service Type. In this way the verifier can also find out which attributes the issuer is trusted to issue.

All the verifier has to do is configure the DNS names of the trust scheme owners that it trusts. When it receives a VC, it extracts the asserted trust schemes made by the issuer in the ToU property, and calls the TRAIN API, passing it the ID of the issuer and the trust scheme it is a member of. The TRAIN API will then check if the VC Issuer is a member of this trust list, and if so return the Service Type URL to the Verifier, so that the verifier can validate that the attributes in the received VC match the schema for the Service Type.

Note that the TRAIN source code can be run by anyone, so there can be multiple distributed copies of this service running on the Internet, and verifiers only need to keep pointers to one or more of them to provide them with backup services (much like the dozen or so root DNS name servers).

Credential Chaining

In the same W3C CCG thread, Daniel Hardman made this point:

I feel like decentralization is running into a difficult tension here: we want to democratize issuance (anyone can do it), but we want to trust a limited set of issuers (or at least, a limited set on any given topic). Anybody can create a COVID test result credential, but we only want to accept them if they were issued by a lab that we have reason to trust. Etc...

One solution to this problem is registries: list trusted sources and have your software check whether the issuer is on approved/accredited list by querying. Of course this re-centralizes around the oracle.

Another solution is chaining: have an accreditation authority issue a VC to issuers, attesting to the issuer's bona fides; verification = verify proximate VC + VC that makes proximate issuer trustworthy. Possibly repeat through several levels of indirection. This still centralizes trust (in the ultimate accreditation parties), but at least it decentralizes the verification.

Right now I am more interested in the chaining model, because I think it is more flexible and it reuses our core primitive (VC verification) instead of substituting a different type of lookup. I also think it scales better. But there can be political reasons to use a registry instead.