Review · October 26, 2026
Anti-Money Laundering Methods Values
This draft defines a single JWT claim, aml, whose value is an array of short strings naming which Anti-Money Laundering (AML) and Countering the Financing of Terrorism (CFT) screening methods were run and passed for the token's subject. It is modeled directly on the amr claim from OpenID Connect, and creates a new IANA registry seeded with five values.
The Problem
Financial regulations in most jurisdictions require sanctions screening, politically exposed persons (PEP) checks, and similar controls before money moves. The authors want a compact way for a token issuer to assert "we ran these screens and they came back clean" so that a downstream party can act on it. The stated use case is KYAPay tokens, the Skyfire draft for agent identity and payment tokens.
How It Fits In
This is a JWT claim registration, not an OAuth mechanism. It depends only on RFC 7519 and the IANA JWT Claims registry. It introduces no new roles, no new endpoints, and no new relationships between client, authorization server, and resource server. The OAuth working group has historically been the home for JWT claim work, so it is in scope in the loose sense, but nothing about it is specific to OAuth.
It is one of a family of six Skyfire drafts from the same authors: the KYAPay Token format, token exchange and usage drafts for it, additional amr values, an ivm (Identity Verification Methods) claim, and this one. The ivm draft is structurally identical to this one with a different vocabulary. The three claim drafts are the building blocks the KYAPay token draft lists as related work.
One thing worth checking: the KYAPay token draft cites this draft only in an appendix listing related work. It does not actually include aml in its defined claim set. So the one named use case does not yet use the claim normatively.
What The Spec Defines
The whole mechanism fits in a paragraph. A JWT issuer MAY include an aml claim containing a JSON array of case-sensitive strings. Each string names a screening method that was performed and succeeded. Values SHOULD come from the new IANA registry, with five initial entries:
ofac: OFAC compliancesanc: sanctions screeningwatch: watchlist screeningpep: politically exposed persons screeningadv: adverse media screening
The registry uses Expert Review with the same experts and mailing list as the amr values registry. The registration template is copied from RFC 8176.
The -01 revision added one substantive sentence: presence of a value means the method "succeeded." Everything else in the diff is formatting.
Interoperability Concerns
The syntax is unambiguous. The semantics are not, and that is where the problems are.
"Succeeded" is undefined. Does sanc mean the screen was run, or that the subject was not on a list? The -01 text says succeeded, which most readers will take as "cleared," but the draft never says so. A screen that ran and produced a hit is also a successful screen from the operator's point of view.
The values overlap. OFAC screening is sanctions screening against one specific set of lists. Watchlist screening is a superset of both. An issuer that ran a commercial screening product could reasonably emit any one, two, or all three of ofac, sanc, and watch for the same check. Two issuers using the same vendor will emit different arrays. A consumer cannot tell the difference between "we checked OFAC only" and "we checked OFAC, EU, and UN consolidated lists."
No time, no screener, no subject scope. The amr analogy breaks down here. OpenID Connect pairs amr with auth_time because knowing an authentication method is useless without knowing when it happened. Sanctions lists change weekly. A sanc value with no timestamp tells the relying party almost nothing it can act on. There is also no way to say who performed the screening or against which party. In an agent payment flow, the subject could be the agent, the person behind the agent, or the counterparty.
Normative references to vendor blog posts. The definitions of four of the five values point to marketing pages from Thomson Reuters, Swift, Moody's, and Persona. These are not stable, not normative, and define nothing precisely. The FATF Recommendations, or the regulatory text itself, would be the obvious anchor. The ofac reference is a Treasury FAQ page.
Compare this to OpenID Connect for Identity Assurance, which already has a verified_claims structure with trust_framework, time, verification_process, and evidence elements. AML screening fits that model well. A flat array of strings is a step backward from what the industry already has for conveying "a regulated check was done."
Mechanisms That Could Be Removed
There is almost nothing to remove. The draft is one claim and one registry. The concern runs the other direction: it is missing the pieces that would make the claim actionable.
The Security and Privacy Considerations sections are each one sentence pointing at RFC 7519. That is not adequate. A claim that asserts "this person is not a sanctioned party" carries regulatory weight and liability. A relying party that waives its own screening because a token said sanc needs to know what trust it is placing in the issuer. The privacy side is also real: PEP status and adverse media results are sensitive personal data, and a pep value with "succeeded" semantics discloses the outcome of that screen.
Writing Quality
The text is clean, uses standard IETF language, and is clearly built from the RFC 8176 template. At ten pages it is mostly boilerplate. The actual content is about one page.
There is one copy-paste leftover: the registration template describes the name as being "for the authentication method or family of closely related authentication methods." The IANA section also cites RFC 5226, which was obsoleted by RFC 8126 in 2017.
The draft is short enough that the lack of justification stands out. There is no explanation of why these five values were chosen, why they are at the granularity they are, or what a relying party is expected to do differently when it sees them.
Authors
Datatracker was unreachable from this session, so the following is based on the drafts in the repo and published history.
- Ankit Agarwal (Skyfire Systems): New to the IETF. Listed on all eleven revisions of the six Skyfire drafts and no other documents. Doing a lot of active writing right now, all in this one family.
- Michael B. Jones (Self-Issued Consulting): Author of dozens of RFCs including JWT, JWS, JWK, JWA, the
amrvalues registry (RFC 8176), Dynamic Client Registration, Authorization Server Metadata, PAR, and DPoP. Co-author on every Skyfire draft and on many other active drafts across OAuth and related groups. The template structure here is his. - Nash Ali (Experian): New to the IETF. Appears on six of the Skyfire draft revisions and nothing else.
Bottom Line
This is a small, well-formed draft that defines something not yet precise enough to interoperate. The syntax is fine. The vocabulary overlaps with itself, the "succeeded" semantics are unstated, and there is no time or provenance. Feedback should push the authors toward either defining what each value actually guarantees and adding a timestamp, or folding this into the existing OpenID Connect for Identity Assurance evidence model where the structure for this already exists. It is also worth asking why the one cited consumer, the KYAPay token draft, does not use the claim.