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

Update review · November 10, 2026

Additional Authentication Method Reference Values Update (-03)

draft-skyfire-oauth-amr-values -03View on Datatracker →

Revision -03 is a substantial rewrite. The -02 draft was a list of one-line definitions. This version adds a composition model, two appendices of examples, and real definitions for most values. Most of the feedback from the previous review was acted on. The tel redefinition and the bg value were not.

What Changed

The draft now registers eleven new values, up from nine.

Added:

  • sna: silent network authentication, where the mobile operator confirms the device holds the SIM for a phone number. Cites the CAMARA Number Verification API.
  • mdc: multi-device credential, meaning a backup-eligible WebAuthn credential. This is the synced passkey signal.
  • uv: user verification in the WebAuthn sense, as opposed to the user presence test that user describes.

Removed:

  • sqa: gone. The history note says kba already covers it.

Redefined:

  • email: now only the channel. What was sent over it is described by code or url.
  • app: defined by what it is, with the product names removed. The issuer omits it when it can't tell what generated a code.
  • facliv: now only the liveness determination. face is the match. Send both when both happened.
  • passkey: now defined as a WebAuthn client-side discoverable credential.

New text:

  • Section 5, "Composing Authentication Method Reference Values". It splits values into channels (email, push, sms, tel) and challenges (code, url), and explains how code differs from otp.
  • Appendix A and B, with examples and a table mapping common terms like "magic link" and "synced passkey" to value sets.
  • Security considerations on email as a channel and on push approval without a transferred secret, both citing NIST SP 800-63B-4.

Was the Previous Feedback Addressed?

Overlap with RFC 8176 values: mostly yes.

  • sqa vs. kba: fixed by removing sqa.
  • facliv vs. face: fixed. The draft says they are not alternatives and gives ["face", "facliv"] as the encoding.
  • passkey vs. hwk/swk: addressed with mdc. A synced passkey is ["passkey", "mdc"], and hwk or swk is added when the issuer knows how the key is stored.
  • app vs. otp: addressed. An authenticator app TOTP is ["app", "otp"].
  • bg vs. geo and rba: acknowledged but not fixed. See below.

code, url, and email mixing delivery with authenticator: yes. The authors kept all three and made the two dimensions explicit. An email code is ["email", "code"]. A magic link is ["email", "url"]. otp is reserved for HOTP and TOTP, and code covers everything else. This is a reasonable answer to the criticism.

tel silently weakened: no. The definition is unchanged from -02. Appendix B now says a voice code is ["tel", "code"] and a call back is ["tel"]. That helps only if every issuer follows a non-normative appendix. A relying party that sees ["tel"] still can't tell which one happened. The security considerations still say nothing about the change.

bg too broad: half. sna gives silent network auth its own value, which is what was asked for. But bg is still registered, and the new text says its other two examples are already covered by geo and rba. After -03, bg has no meaning that another value doesn't carry better. It should be removed.

Each value needs more than one line: yes. This was the main writing complaint and it's resolved for most values. code and url are still one-liners, but Section 5 covers them.

Could this skip the working group and go to the designated experts: not addressed. The draft is still Standards Track and still updates RFC 8176 because of tel.

New Concerns in -03

The composition rules are not normative. Section 5 says an issuer "will typically return both" a channel and a challenge. Appendix B is explicitly non-normative. The stated purpose of Appendix B is "so that implementers of the same arrangement emit the same set of values." That's the interoperability requirement of the whole document, and it's stated as a hope. For SMS the draft blesses two encodings outright: ["sms", "code"] and ["sms"] are both correct.

push is defined two ways. Section 5 calls it a channel and says "a channel value alone does not say what the person did." Appendix B says push approval is ["push"] alone. So push alone does say what the person did. Section 3.7 also defines it as an "out-of-band authentication interaction," which is more than a channel.

app and push still collide. A push approval in Okta Verify is a push to an authenticator app. The app definition says to include it whenever the issuer knows an authenticator app was used. Appendix B says the encoding is ["push"]. Is it ["push"] or ["app", "push"]? The -02 version of this question was app vs. otp vs. push. Only the otp half was answered.

A relying party can't require a device-bound passkey. mdc marks a credential as syncable. The draft then says "the absence of this value does not indicate that a credential is a single-device credential." Appendix B lists ["passkey", "hwk"] for a device-bound passkey, but a synced passkey from an issuer that doesn't emit mdc can produce the same array. Policy like "no synced credentials for admin access" needs a positive signal for device-bound, and there isn't one.

passkey leaves out non-discoverable credentials. It's defined as a client-side discoverable credential. A security key used as a second factor with a server-side credential gets no new value and falls back to hwk. That may be intended. The draft should say so.

uv isn't used in any example. It's a useful addition, since user only ever meant presence. But no entry in Appendix B includes it, including the passkey rows where it matters most. The draft also doesn't say how uv relates to the existing pin, fpt, and face values. Does an issuer that knows the modality send ["uv", "fpt"] or just ["fpt"]?

bg and sna have no combination rule. sna is described as a subset of bg. Appendix B shows ["sna"] alone. Nothing says whether ["bg", "sna"] is expected, allowed, or redundant.

Authors

No changes to the author list. Max Gerber was added to the acknowledgments.

Bottom Line

This is a much better draft than -02. The authors removed the duplicate value, separated channels from challenges, and wrote down how the new values combine with the RFC 8176 ones. The passkey story is now usable with mdc and uv.

Three things are left to push on. Remove bg, since the draft's own text shows sna, geo, and rba cover it. Stop redefining tel, or at least add security considerations explaining that existing relying parties will see a weaker meaning. Make the composition rules in Section 5 and Appendix B normative, and fix the push and app inconsistencies while doing it. Without that last change, two identity providers can still describe the same login with different arrays.