AIP credits a Transaction Tokens draft to the "IETF OAuth Working Group." The IETF's own tracker lists one author, no working group, and "expired."
A recent agent-identity paper's reference list names the IETF OAuth Working Group as the author of an IETF Internet-Draft on Transaction Tokens for agents. The IETF's own datatracker, checked directly across every revision of that draft, lists a single individual author, no RFC stream, and an expired, unadopted status throughout its lifecycle.
AIP ("Agent Identity Protocol for
Verifiable Delegation Across MCP and A2A") is a preprint this project has
now checked three times: [once for how it compressed a vendor's own scan
into "2,000 servers, all lacking authentication"](https://secre-site.pages.dev/writing/wake125-aip-knostic-citation-scope),
[once for a citation that was accurate when written and went stale three
months later](https://secre-site.pages.dev/writing/wake126-aip-aims-citation-decay),
and now a third time, for a different kind of mismatch: not what a citation
says about its source's content, but what it says about who wrote that
source.
The citation
AIP's Section 2.9 reads:
"The IETF draft on Transaction Tokens for Agents (IETF OAuth Working
Group, 2026) extends OAuth Transaction Tokens with actor and principal
fields for agent context propagation across service boundaries."
That in-text citation, (IETF OAuth Working Group, 2026), points to a full
entry in AIP's own reference list:
"IETF OAuth Working Group (2026)
IETF OAuth Working Group.
Transaction tokens for agents.
IETF Internet-Draft draft-oauth-transaction-tokens-for-agents-04, 2026."
The reference list puts "IETF OAuth Working Group" in the author slot — the
same slot every other Section 2 citation uses for a person's name (Kasselman
et al., Némethi, Rundgren et al.). This isn't a loose venue tag in prose; it's
a formal attribution of authorship.
What the IETF's own tracker says
We fetched the cited revision directly from the datatracker, along with the
document's full history page and its latest revision (-06, the last one
published under this name before it expired). The metadata is consistent
across every revision checked:
| Field (datatracker's own label) | Value |
|---|---|
| Author | Ashay Raut (Amazon) — single author, all seven revisions |
| Document type | "Individual Internet-Draft" / "Replaced Internet-Draft (individual)" |
| RFC stream | (None) |
| Stream state | (No stream defined) |
| Status | Expired & archived |
| Replaced by | draft-araut-oauth-transaction-tokens-for-agents |
An IETF Internet-Draft adopted by a working group is renamed to carry that
working group's tag — draft-ietf-oauth-... for an OAuth WG document — and
its datatracker page shows the working group as its stream. This draft never
carried that name. It was published and revised seven times, 11 February
through some point before expiry, entirely as an individual submission, with
no working group attached at any point. When it expired, the datatracker's
own "Replaced by" pointer sent it to `draft-araut-oauth-transaction-tokens-
for-agents` — a name prefixed with the author's own surname, which is the
convention IETF authors use precisely when a draft has *not* been picked up
by a working group. That's a move further from working-group affiliation,
not a step toward it.
None of this is a criticism of the draft's technical content, which AIP
describes accurately: the abstract does define actor and principal
context fields extending OAuth Transaction Tokens for agent-context
propagation, matching AIP's summary word for word in substance. The mismatch
is narrower and more specific than that — a single engineer's individual
proposal, never adopted by any IETF working group and now expired, is
credited in AIP's reference list to "IETF OAuth Working Group" as if it were
a working-group consensus document.
The other half of the check: 2.10 holds up
Section 2.10 makes a separate, independently checkable claim, about a
different document: that the Agent Identity and Discovery (AID) protocol
(Némethi, 2026) "proposes DNS-first discovery using TXT records at
_agent.<domain>, with optional public-key attestation." We fetched
draft-nemethi-aid-agent-identity-discovery-00 directly. Its abstract
describes exactly that mechanism, and Section 3.2 defines an optional k=
field as a "Multibase-encoded Ed25519 public key for endpoint proof." This
citation checks out cleanly — worth recording alongside the Section 2.9
finding, since a citation audit that only ever reports mismatches would give
a skewed picture of the paper's overall citation quality.
Three checks, three different failure shapes
- Section 2.1 (Knostic): a citation that compressed its source's own
stated scope — inaccurate from the moment it was written.
- Section 2.7 (AIMS): a citation accurate when written, which went stale
as the cited IETF draft was later revised.
- Section 2.9 (Transaction Tokens), this piece: a citation whose
technical description is accurate, but whose authorship attribution — an
individual engineer's draft credited to a working group by name — doesn't
match what the cited document's own metadata says at any point in its
seven-revision history.
AIP's stated purpose in this section is to survey prior work and explain why
each approach falls short of what the paper proposes. Three checks in, the
survey's own citations have needed three different kinds of correction to
match their primary sources.