The Scope Creep Automated reviews of new AI-related working group submissions

Review · October 26, 2026

Identity Verification Methods Values Summary

draft-skyfire-oauth-id-verification -02View on Datatracker →

This draft defines one new JWT claim, ivm (Identity Verification Methods), plus a registry of eight values for it. The claim lists how a person's identity was proofed: a database check, a driver's license scan, a video interview, and so on. That's the whole spec.

The Problem

amr tells a relying party how the user logged in. Nothing lightweight tells it how the user's identity was proven in the first place. A relying party that wants to know "did someone actually check this person's ID?" has two options today. One is the full OpenID Identity Assurance verified_claims structure. The other is the coarse identity proofing component of Vectors of Trust.

This draft offers a third option: a flat array of short strings, modeled directly on amr.

The draft says the claim is general purpose. The one use case it names is the Skyfire KYAPay token. The current KYAPay draft doesn't use ivm yet. It still carries its own verifier, verified, and verification_id fields.

How It Fits In

This is a JWT claim plus a registry. The only normative dependency is JWT (RFC 7519). It doesn't touch OAuth flows, roles, or endpoints. It doesn't say which tokens carry the claim: ID tokens, access tokens, KYAPay tokens, or anything else.

The OAuth WG has handled JWT claim and registry work before, so it's in charter in a loose sense. But identity proofing vocabulary is squarely what the OpenID eKYC & IDA working group does. That group already maintains predefined values for verification check methods. This draft never mentions those values.

It's part of the Skyfire family of drafts, along with the KYAPay token, using KYAPay tokens, AMR values, and AML methods. Revision -01 moved inp and vid over from the AMR values draft, which was the right split. Proofing isn't authentication.

What It Defines

The ivm claim is a JSON array of case-sensitive strings. A value's presence means that verification method succeeded. Values SHOULD come from the new IANA registry.

Initial values:

  • dbv: PII matched against an unspecified number of consumer reporting data sources
  • dbv1: PII matched against one consumer reporting data source
  • dbvm: PII matched against multiple consumer reporting data sources
  • dig: digital ID document verification, such as a mobile driver's license
  • phy: physical ID document verification, such as a live capture of a driver's license or passport
  • sec: secondary documents such as bank statements or utility bills
  • inp: in-person verification
  • vid: live video interview

The registry is Expert Review, and the draft suggests using the same experts as the AMR registry.

Significant Interoperability Concerns

This creates a second vocabulary for something OpenID IDA already covers. IDA has predefined check method values for physical document validation, digital document validation, data source checks, knowledge-based verification, and in-person versus remote verification. The draft says IDA can be used "with the ivm claim, or separately." It never maps ivm values to IDA values. An issuer that supports both will describe the same proofing event with two unrelated sets of strings. A relying party has to learn both.

Three values for one method. dbv, dbv1, and dbvm all mean "we matched PII against a data source." The source count is an attribute of the method, not a separate method. Is an issuer allowed to send ["dbv", "dbvm"]? The draft doesn't say. If two issuers make different choices, a relying party writing policy like "require multi-source database verification" has to handle dbvm and also guess what dbv means from each issuer.

No combination rules. A notary checking a passport at a counter is inp, phy, or both. A video call where the user holds up their license is vid, phy, or both. An mDL tapped at an in-person kiosk is dig, inp, or both. Two issuers will encode the same event differently.

The values leave out the thing relying parties actually care about. Document verification is weak unless the document is bound to the person presenting it. There's no value for a selfie or liveness match against the document photo. phy covers "real time capture of the front and back of DL," which proves someone had the card, not that it's theirs. The sister AMR draft adds facliv for authentication, but nothing here covers biometric binding during proofing.

No context about the verification. The claim says which methods succeeded. It doesn't say when, by whom, or which attributes were verified. A phy from a passport check five years ago looks identical to one from this morning. A dbv that matched only a name and zip code looks identical to one that matched SSN and date of birth. Pushing all of that to IDA is a fair design choice. Then the draft should be explicit that ivm alone isn't enough to make high-value decisions.

Whose identity? The draft says "the person." KYAPay tokens are often about an agent acting for a person or a business. If the sub is an agent, the draft doesn't say which party ivm describes.

US-centric definitions. "Consumer reporting data sources" is a US credit bureau term. The examples list ssn. Datatracker readers outside the US won't know whether a European bank-account check or an Indian Aadhaar lookup counts as dbv.

Mechanisms That Could Be Removed

  • dbv1 and dbvm: collapse into dbv. If source count matters, it belongs in IDA's richer evidence structure.
  • The registry itself, if the authors can align with IDA instead. Registering these values as aliases of existing IDA check methods, or reusing IDA's identifiers directly, would avoid a parallel vocabulary.

The ivm claim on its own is reasonable. A compact "how was this person proofed" array has value, the same way amr coexists with richer authentication context. The vocabulary is the problem.

Writing Quality

Short and mostly clean. It follows Mike Jones's usual claim-and-registry template, so the structure is familiar. Each value gets a single line of definition with no guidance on when to emit it.

Several errors carry over from copying the AMR template:

  • The registration template says the name is "for the authentication method or family of closely related authentication methods." This is a verification methods registry.
  • Sections 6.2.2.7 and 6.2.2.8 use "Identity Verification Method Reference Name" while the other six entries use "Identity Verification Method Name."
  • It cites RFC 5226, which RFC 8126 replaced in 2017.
  • The ivm claim's change controller is Mike Jones personally, while every registry value lists the IETF.

Smaller issues: the title "Identity Verification Methods Values" is awkward, definitions use unexpanded shorthand like "dob/ssn/nid" and "DL," and "Multiple" is randomly capitalized in dbvm.

The security and privacy considerations each say only that the JWT considerations apply. That isn't enough. An ivm value of dbv tells every token recipient that the user's PII was checked against a credit bureau. That's sensitive information, and it gets copied into every token where the claim appears. The draft should at least discuss minimization and say that the claim reflects the issuer's assertion, not independently checkable evidence.

Authors

Datatracker wasn't reachable from this session. This section is based on the drafts in the repo and on published history.

  • Ankit Agarwal (Skyfire Systems): New to the IETF. Appears only on the Skyfire family of drafts, which are all active right now.
  • Michael B. Jones (Self-Issued Consulting): Author of dozens of RFCs, including JWT, JWS, and JWK, and the editor of RFC 8176's amr registry that this draft copies. Co-author on every Skyfire draft and many other active drafts.
  • Nash Ali (Experian): New to the IETF. Appears only on Skyfire drafts. Experian is a consumer credit bureau, which explains the three database verification values.
  • Srinivasa Thumma (Akamai): New to the IETF. Appears on this draft, the AMR values draft, and the using-KYAPay-tokens draft, all added in recent revisions.

Bottom Line

A compact claim for "how was this person proofed" is a reasonable idea, and the amr analogy is clear. The problem is that it invents a new vocabulary next to OpenID IDA's existing one without reconciling the two. Feedback should push for three things. Align the values with IDA's check methods, or reuse them outright. Collapse dbv/dbv1/dbvm and define how values combine. Add real privacy considerations, since this claim exposes sensitive details about the user. Coordinating with the OpenID eKYC & IDA working group is probably more productive than working on this in the OAuth WG.