DKIM2 draft tracker: where every draft and open question stands
This page tracks the IETF record for DKIM2. It is edited in place, and the changelog says what moved. If you are new to the protocol, start with DKIM2 explained. Dates are 2026 and UTC unless a year is shown.
Where does DKIM2 stand?
Nine DKIM2 drafts are active: three working-group documents and six individual submissions, which have no formal standing until a working group adopts them.
- draft-ietf-dkim-dkim2-spec (-09, dated 1 November 2026): working group; Clayton, Chuang, Gondwana. Adopted 18 March (call closed 16 March); first revision 24 March. Revisions -07 to -09 each changed requirements. Next: WGLC after the 18 November meeting, unless significant issues remain.
- draft-ietf-dkim-dkim2-dns (-03, 2 Nov 2026): working group; Chuang, Clayton. Adopted 17 June (filename 20 July), replacing draft-chuang-dkim2-dns; retitled “Domain Name Specification for DKIM1 and DKIM2”; updates RFC 6376 if approved. Clayton calls it “pretty final IMO”, with “limited discussion” of its effect on DKIM1. Next: the 18 November agenda.
- draft-ietf-dkim-dkim2-bcp (-01, 9 Sep): working group; Herr. Adopted 17 June (filename 18 June), replacing draft-herr-dkim2-bcp. It predates the last three spec revisions; there is no newer one.
- draft-chuang-dkim2-sender-policy (-04, 19 Oct): individual, Experimental; Chuang, Gondwana. Its header names the DKIM working group since -04, but the group has not adopted it and has issued no adoption call.
- draft-clayton-dkim-canmodify (-00, 2 Nov): individual, Standards Track; Clayton. Next: ten minutes on the 18 November agenda.
- draft-moccia-dkim2-deployment-profile (-08, 30 Oct): individual, Informational; Moccia. Gives the milter profile (a milter is a plug-in that mail servers call to inspect or change a message) its own header names, such as
DKIM2-Chain-SigandDKIM2-Chain-mf, apart from the spec’sDKIM2-Signature. - draft-chuang-dkim2-localized-header-fields (-00, 9 Jun): individual, Experimental; Chuang.
- draft-gondwana-dkim2-authres (-02, 2 Nov): individual, Standards Track; Gondwana.
- draft-gondwana-dkim2-debug-header (-03, 2 Nov): individual, Informational; Gondwana. Defines an
X-DKIM2-Infoheader for interop testing; “unlikely to be published” (abstract), and “Nothing in this document is normative” (introduction). - DMARC reports for DKIM2: a proposal on the list; no draft exists. It needs one for the DMARC working group.
- draft-ietf-dmarc-arc-to-historic: -01, posted 21 October 2026. Requests Historic; RFC 8617 is still Experimental. The DMARC working group’s charter deadline, 16 October 2026, passed without a Working Group Last Call. Next: DMARC working group milestone, due 30 November 2026. More in our piece on ARC email authentication.
- DKIM charter milestone: adoption milestones done. Next: send all documents to the Internet Engineering Steering Group (IESG, which approves IETF standards) by 31 March 2027.
Expired or replaced drafts people still cite:
- draft-ietf-dkim-dkim2-motivation (-02, 3 Nov 2025): working group, adopted July 2025, expired 7 May. Gondwana, Clayton, Chuang.
- draft-gondwana-dkim2-modification-alegbra (-04, 17 Oct 2025): replaced by draft-gondwana-dkim2-mailversion, a Dead working-group document since 2 April, merged into the spec. The mailversion draft expired 7 May. The datatracker misspells the slug
alegbra, and the link copies it because any other spelling returns a 404. - draft-robinson-dkim2-message-examples (-00, 30 Jun 2025): expired 1 January. By Allen Robinson at Google and Tobias Herkula at 1&1.
draft-ietf-dkim-dkim2-header(-00, 3 Nov 2025): merged into the spec, dead since 17 April, expired 7 May. Gondwana, Clayton, Chuang.- Allen Robinson’s bounce-processing draft (-01) and R. Latimer’s recipient and next-domain signing draft (-00) expired in January and May.
Two DKIM1 drafts concern signing algorithms, not DKIM2, and neither is adopted. draft-xie-dkim-sm2-sm3-00 (30 October) would update RFC 6376 to add an SM2-SM3 algorithm. draft-buck-dkim-pq-dkim1-00 (2 November) expects a post-quantum signature in a separate DKIM-Signature header beside a traditional one; it defines no algorithm identifiers and does not update RFC 6376.
Is the best-practices draft current?
Only in part. It predates the last three spec revisions and the DNS rewrite. Its §3.5 still recommends t=y test-mode selectors, which the DNS draft deprecates (§4), and its §5.5 still tells forwarders to honor donotmodify and donotexplode, which spec-07 removed. Read it for practices, not flags.
It is looser on timestamps. The spec says a message is not Fully Verified (its term for a message whose whole signature chain checks out) if any signature is over 14 days old (§14.3); the draft says verifiers “are permitted to treat” such a signature “with suspicion” (§3.6), citing older spec section numbers. On key rotation, §3.5 asks signers to publish new keys under a new selector, keep the old key published until in-flight mail has cleared, and never reuse a retired selector name. It gives no number of days; on our reading the outer bound is the spec’s 14-day window.
What is settled?
The design’s spine has held through every working-group revision. These five commitments are in the current spec, and the open questions below do not reopen them.
- Per-hop signing. Every system that handles a message adds a
DKIM2-Signatureheader. The spec calls the result a Chain of Custody that “assists Verifiers in distinguishing between messages that were intended to be sent to a particular email address and those that are being ‘replayed’ to that address” (draft-ietf-dkim-dkim2-spec-09§1). - Envelope binding. Each signature records the SMTP envelope, the
MAIL FROM(sender) andRCPT TO(recipient) addresses, asmf=andrt=. The receiver must match them exactly against the highest-numbered signature (spec §10.5, §10.6, §12.2; the verifier’s check is §14.4). A hop inside one system may carrynd=, naming the next signing domain, but the run must end with a signature that carries both (§10.7). - Chained
i=indices. The originator usesi=1, and each later hop adds one. A message with gaps or duplicates “is not a Fully Verified DKIM2 Message” (§10.1). - Replay resistance. A message is not Fully Verified if any signature is over 14 days old when accepted over SMTP, or dated more than a day ahead (§14.3). The motivation draft adds: “DKIM2 headers will always have timestamps so that ‘old’ signatures have no value” (§3.1).
- DNS reuse. Keys are TXT records in the
_domainkeynamespace from RFC 6376, with no new record type; they “are no different, and are stored in the same locations as those for DKIM1” (§3.7).
What changes inside the key record?
The DNS draft changes what goes inside the record, not where it lives:
p=format. An RSAp=value may be a bare RSAPublicKey or a SubjectPublicKeyInfo structure (DNS draft §4.2), settling an ambiguity raised by RFC 6376 errata 3017, 6674 and 7001. Existing records stay valid.- Deprecated tags.
h=,n=,s=andt=“SHOULD NOT be used in DKIM1 key records, and DKIM2 ignores them” (§4), as an update to RFC 6376. A selector witht=ygets no testing-mode cover from a DKIM2 verifier, though DKIM1 verifiers still honor it. This is a draft proposal, not in force. - One record per selector (§3).
What is the Domain Policy Record?
Spec §3.8 and DNS draft §5 add a second kind of TXT record at _domainkey.example.com itself: a tag=value list that must begin v=DKIM1 and may carry one flag, f=strict. It applies only to the exact domain that publishes it, and DKIM1 ignores it.
It concerns what counts as Authentic: a From: domain that relaxed-matches the d= of the right signature (§11). Since -09 the match also runs upward, so an email service provider (ESP) signing as d=em.example.com can make a From: at example.com Authentic (§11.1). With f=strict the upward match stops and a verifier reports a FAIL, “From … forbids subdomain signing” (§14.8); without the record, the tag or the flag, subdomains stay authorized.
The Security Considerations (§20) say a domain whose customers or tenants publish DNS records beneath it needs f=strict, and call a control that defaults to “authorized” unfortunate but the only way to avoid breaking existing practice. A temporary DNS failure is a TEMPERROR; more than one v=DKIM1 record is a PERMERROR. Do not publish it yet: it is draft text in two documents that are still moving.
What is still being argued?
Two terms first. A Message-Instance header records one version of the message. A recipe, kept inside it, lists the edits a forwarder made so a verifier can undo them and recheck earlier signatures.
Which way should recipes be numbered?
Bron Gondwana (Fastmail) defends bottom-up numbering, so a recipe can read Authentication-Results: [1, 3] rather than [2, 4]. Hannah Stern (1&1 Mail & Media) asked on 17 March why not top-down throughout and preferred it so hashing and recipe execution stay consistent. Stern agreed the draft as written means bottom-up.
Where it stands: both positions are on the list; the draft numbers bottom-up.
In what order is the signing input built?
In his 14 March hackathon report, which he says was generated by two copies of Claude Code, with some input from him, Gondwana argued for interleaving: when signature i=1 was created, only Message-Instance 1 existed. Stern defended the wording already in §11.5 of the earlier Clayton spec: all Message-Instance headers first, then all signatures.
Where it stands: spec-09 §12.8 still puts all Message-Instance headers before all signatures.
Should JSON sit inside base64?
Early drafts put JSON inside base64 inside tag values. The first working-group revision, -00 (24 March), records “Removed JSON for hashes, signatures and SMTP parameters.” JSON survives inside recipes, where the list reopened it from 8 to 17 September as “base64+JSON on the critical path”, with Vittorio Moccia, Gondwana and Stern posting. Spec-07 took one change from that thread, outlawing duplicate JSON keys.
Where it stands: gone from signature tags, still in recipes in spec-09.
Does rt= leak recipients?
During the October 2025 call to adopt an earlier header draft, R. Latimer (Inveigle.net) objected on 13 October 2025 that rt= risks data leakage that header-only construction cannot fully remove. The dispute concerns recipient privacy, not whether the envelope is bound. The answer so far is operational: the tag stays, and the best-practices draft’s §5.4 concedes that a forwarder relaying to a subscriber’s real address “will, as a side effect, expose that address” and says such forwarders SHOULD deliver to “a per-subscriber, per-forward alias local part”.
Where it stands: the tag stays; mitigation is best-practices guidance.
Where should DKIM2 policy live?
The working group first declined to decide. The IETF 126 minutes say the core spec “will remain silent on From alignment”, with guidance going to the best-practices draft “and perhaps make updates to DMARC”. That draft (§7.7) calls it “an open architectural question that this document does not attempt to resolve.”
Chuang forced the issue on 26 August, proposing to split per-message constraints (DKIM2 signature flags) from domain-level enforcement (the DMARC record). Todd Herr, the best-practices author, replied within hours: “Unless and until there’s an updated RFC 9989 published that acknowledges DKIM2 as an Authenticated Identifier, there is no world where DMARC alignment validation is applied corresponding to a DKIM2 signature.” John Levine agreed on 2 September that policy belongs in a separate draft, and called the proposed unaligned flag redundant, since a receiver can compare d= with the From: domain.
One half then reached the spec. Richard Clayton proposed that a From: domain must match the d= of the signature added when that From: was introduced or changed. Tobias Herkula (1&1 Mail & Media) wants it “added directly to the DKIM2 specification”, and Herr backed language saying “the DKIM2 signing domain MUST match the RFC5322.From domain whenever the domain is introduced or changed in a message.” Spec-07 wrote it in (§11): signers MUST pick such a domain (§12.4), and a verifier reports a mismatch as a FAIL (§14.8; a PERMERROR in -07).
The other half, DMARC itself, is where Herr objected on 17 September. In the same message he wrote that he believed “adding DKIM2 as an authentication mechanism for DMARC will slow, rather than accelerate, adoption of DKIM2, because a DMARC pass will still be achievable with just SPF, or just DKIM.” The spec says nothing about DMARC policy. The DKIM charter says the group “will also liaise with the DMARC working group on adding its specification as a new recognized authentication mechanism.” It adds: “If the DMARC working group has concluded by that time, this working group can undertake that update in coordination with the responsible Area Directors.”
Where it stands: the spec settles From: against d=; DMARC’s role is open.
Must a body change come with a From rewrite?
On 12 November, Richard Clayton set out a way forward: “It is suggested that a way forward … is to make it a MUST that ANY body modification MUST be accompanied by a change to the From header field.” In his words, “Essentially we make From-munging compulsory if you change the body.”
The same day, before Clayton’s message, Pete Resnick, a co-chair writing as a participant, had argued against making From rewriting the default answer: if senders may say donotmodify and receivers “blindly follow that instruction”, most, if not all, senders will, and From munging breaks replies and address books. Clayton’s message replies to him. Gondwana wrote on 11 November that he would love to talk about From rewriting and when to do it.
The background is spec-07’s removal of donotmodify and donotexplode.
Clayton wrote on 20 September that a “strong case has been made” for removing donotexplode. On 26 September he saw “a strong case for removing ‘donotmodify’ from the DKIM2 spec itself”, and Gondwana replied that removing it “seems right to me”. His spec-07 announcement says there appeared to be “WG rough consensus”. On 2 November he posted the CanModify draft, where the originator lists the changes it will accept: none, bodywrap, subjectprefix or alterfrom. It is valid only on a Fully Verified DKIM2 Message, only if added at origination, and honoring it is local policy (§3).
Where it stands: a proposal under discussion; the spec text has not changed.
How would DMARC reports cover DKIM2?
On 11 November Chuang proposed “modifying RFC9990 DMARC reports to support DKIM2”: a result per signature and selector, with the spec’s feedback and feedhere flags reaching forwarders. Levine called it “quite reasonable” but said it “has process problems because this WG is chartered to update DKIM but not DMARC”. The fix is a draft for the DMARC working group.
Where it stands: a proposal; no draft exists.
Which post-quantum algorithm should DKIM2 use?
Chuang opened the question on 19 June with a threat-model statement: “Google feels a PQC migration is necessary by 2029.” Stern answered on 14 September: “I’d be strongly for standardizing only a hybrid, not a ‘pure’ pqc signature algorithm.” On 5 October Chuang reported that at the 23 September interim “the direction is to use a composite aka hybrid: ML-DSA-44 / Ed25519.” He added: “This would apply only to DKIM2, skipping DKIM1.” What followed:
- DNS draft -01 wrote in the composite. Signers and verifiers MUST implement
rsa-sha256and SHOULD implemented25519-sha256andmldsa44-ed25519-sha512; both SHOULDs become MUSTs “upon the completion of the DKIM2 specification” (§3.3). The composite pairs ML-DSA-44, the lattice signature NIST standardized in FIPS 204, with Ed25519 as “backstop protection for ML-DSA in case it has a latent undiscovered vulnerability”. It came from the LAMPS (Limited Additional Mechanisms for PKIX and SMIME) working group’sdraft-ietf-lamps-pq-composite-sigs, which the IESG approved in April and which was published as RFC 10073 on 6 November. Its public key was 1,344 bytes (1,792 base64 characters inp=, against 392 for 2048-bit RSA) and each signature 2,484 bytes. - Objections came within a week of the 5 October report, before -01 posted. Chuang reported on 10 October, after checking with OpenSSL’s Viktor Dukhovni, that composite support in OpenSSL “sounds like it will not be supported in the near term”. Moccia argued for two separate signatures: “Composite fails closed, since there is nothing they can verify. Two signatures fail open, since the classical signature passes.” Levine agreed that “if we’re going to do PQ, MLDSA44 is currently the best bet”, but called it “more important to support multiple algorithms”.
- The draft’s own author moved. Within 75 minutes of -01 posting, Chuang was “worried that the downsides outweigh the security improvements it brings”. Without library support “email operators will have to implement their own composite implementation”, so he proposed “RSA, Ed25519, and ML-DSA-44 as they all have native support in the major crypto libraries”.
- The spec followed the interim, then reversed. Spec-07 added the composite. Clayton wrote in his announcement that he did it “as instructed by the interim (but I now think this is not the right thing to do, see another message of mine)”, and in that message proposed “Use MLDSA-44 rather than a hybrid.” DNS draft -02 kept the composite and added SHA-512 variants of RSA and Ed25519, but spec-08 replaced it with plain ML-DSA-44 (§3.4) and added AND/OR rules for combining signatures, “to allow phased deployment of new algorithms alongside ‘hybrid’ schemes” (§14.6). On our reading, that gives Moccia’s two-signature route a place in the spec.
- The DNS draft caught up.
draft-ietf-dkim-dkim2-dns-03listsrsa,ed25519andmldsa44and drops the composite and the SHA-512 variants. DKIM1 signers and verifiers MUST supportrsa; DKIM2 verifiers MUST support all three (§4.2). The spec requires verifiers to implement all five hash and signing algorithms (sha256,sha512,rsa-sha256,ed25519-sha256,mldsa44; §3).
Neither draft drops RSA. Allen Robinson gave the reason on 2 October: “we’ve kept RSA signatures in DKIM2 so that service providers don’t need to chase down their customers to get new DNS records deployed.”
Where it stands: both drafts list plain ML-DSA-44 and keep RSA.
Can DNS carry a post-quantum key?
Not comfortably. Steve Atkins (Word to the Wise) noted on 10 November that an ML-DSA-44 public key is 1,312 bytes, “base64 encoded about 1750 bytes”, so validating it needs DNS over TCP: the answer will not fit a UDP packet, “usually configured to be 1220 bytes”. The DNS draft agrees on the key size (§4.2).
Word to the Wise measured the public key as published in DNS: 2048-bit RSA 460 bytes, ed25519 147 bytes and mldsa44 1,861 bytes. DNS Flag Day 2020 recommends a 1,232-byte EDNS buffer.
Levine agreed (“Yup.”). The DNS Operations working group, he wrote, is having a similar discussion about post-quantum DNSSEC signatures: “There is considerable disagreement about how well DNS over TCP works.” Jim Fenton had mentioned it at IETF 126. None of the three working-group drafts mentions TCP or UDP. Only the individual deployment-profile draft (-08) says so: the lookup “is answered with a truncated response and repeated over TCP”.
Where it stands: open; only an individual draft addresses TCP.
What does the sender-policy draft propose?
Chuang posted draft-chuang-dkim2-sender-policy on 10 September and revised it three times in eleven days; -04 followed on 19 October with Gondwana as co-author. It is the first draft to spell out how RFC 9989 and DKIM2’s forwarding model would fit together (section numbers are -04’s). It makes four changes to DMARC and one to DKIM2:
- An
auth=tag in the DMARC record (§2.2.1) lists the methods a domain supports “as both an SMTP MTA sender and receiver”:dkim,dkim2,spf. It is “completely optional”; if absent “relying parties MAY assume that both DKIM and SPF are supported” but not DKIM2, and a domain may support DKIM2 without it “during the bring-up period”. Since -02, a message with no declared method means “the mail receiver SHOULD apply the sender’s enforcement policy.” - The Author Domain comes from
i=1(§2.1). Receivers SHOULD use the first signature’sFrom:(the RFC 5598 author), not the newest. Through -01 this was a MUST with a fallback to the newestFrom:; from -02 it is a SHOULD with no fallback, after Chuang said on 19 September he would “remove support for i=n”. - DKIM2 becomes a DMARC method (§2, §2.2.2), modifying RFC 9989 §4.3. A validated chain is a DMARC “pass”, failing DKIM2 headers are “fail”, and none is “neutral”. Alignment is relaxed, between the Author Domain and the
d=of thei=1signature. p=andsp=are generalized (§2.1). The values stayreject,quarantineandnonebut apply “for any failing domain owner’s constraint/validation subject to local policy, and not just for alignment.”- Per-message policy moves into DKIM2 flags (§3).
donotmodifyanddonotexplodeconstrain downstream relays,explodedis a report, and -04 droppedunaligned. A receiver that will not honor a constraint “MUST NOT relay the message outside of its control.”feedbackandfeedhereare marked “TBD”.
The practical piece is §3.2, for relays without DKIM2, and -04 turned it around. Through -03 a sending relay checked the next hop’s auth= and could bounce, strip the DKIM2 headers or end the chain with an empty rt=. Gondwana proposed the opposite on 1 October: the sender does nothing new, and the final receiver looks up a DNS record on the intermediate’s domain. David Conradie suggested the same day that auth= could carry that signal, and Gondwana agreed. Chuang announced -04: “I’ve also removed the ‘unaligned’ flag proposal. Comments are very much welcome.”
Under -04, a receiver whose chain fails MAY look up the DMARC records of the newest signature’s rt= domains; if one does not list dkim2 in auth=, it may treat the failure as a forwarder without DKIM2, override it and report an RFC 8601 “policy” result (§3.2; example §4.2.4). A receiver that forwards such a message “MUST NOT append a new DKIM2 signature over the existing broken chain”: it deletes every earlier DKIM2-Signature or forwards without signing.
Five caveats before you treat it as a roadmap:
- No Security Considerations or IANA Considerations sections, though it mints a DMARC tag that would need registration.
auth=protects only domains that opt in. Absence means “assume DKIM and SPF”, so stripped DKIM2 stays invisible on a domain that never published the tag. On our reading, §3.2 extends that gap to replay.- The
i=1preference has a deliverability cost the draft does not analyze. One broken link makes the DKIM2 result “fail” (§2.2.2), and the §3.2 override covers one kind of break, at receivers that apply it. - It targets a spec that has moved. It still defines policy for
donotexplodeanddonotmodify(§3.1.2, §3.1.3), every spec section it cites has been renumbered, and forwarding a broken chain unsigned is what spec §13.1 says “MUST NOT be sent to another system (without some local arrangement being in place)”. - Declaring
dkim2is all or nothing per domain. In the thread about a DNS-record policy, Robinson pointed out that “for domains involving more than one mail provider (ex. separate mailbox and bulk mail providers), DKIM2 support can only be declared after all of them support DKIM2.” On our reading, once a domain listsdkim2, receivers can no longer excuse a broken chain on its mail as a legacy hop.
Herr’s 10 September point, that DMARC alignment cannot use DKIM2 unless RFC 9989 §4.3 is updated, was met in -01, and Clayton, after questioning the tag, concluded on 16 September that “the DMARC record is exactly the place to put auth=dkim2”. But the DMARC working group’s charter says it “will adopt no other documents or work items” beyond ARC to Historic, so treat the draft as a watchlist item and do not publish auth=.
Is anyone running DKIM2?
Implementers have tested since the IETF 124 hackathon. The IETF 124 minutes record that Allen’s verifier validated all of Bron’s email files and the reverse, and that Richard Clayton implemented everything from scratch. By IETF 126, the minutes record Go implementations from Gondwana and Atkins and a Rust one in the mail-auth library and Stalwart server, cross-tested in a Docker matrix, with a weekly interop call; the only minor issue was message folding.
On 29 September, Gondwana told the list that a message “was DKIM2 signed on the way out from Fastmail.” Fastmail signs for a handful of domains and validates DKIM2 on inbound mail for all its customers, using the Perl library from the interop repository. It is the only mailbox provider to have said so on the list. Asked whether the format is stable enough, Gondwana answered: “Meh, code can be changed.” He added: “DKIM2 headers have value for 2 weeks max, so it’s not a big deal if/when the format changes - the old messages will just have old headers.”
The spec has been revised 9 times since the March adoption call. Spec-07 removed two flags, spec-08 swapped the post-quantum algorithm, and spec-09 added the Domain Policy Record and the NONE result. A list message Wei Chuang sent from google.com on 17 October carried a DKIM2-Signature; Google has not said it signs with DKIM2.
Where is the working group in its process?
On 21 October the working group chair updated the charter milestones. Adopting the implementation-guide and mechanism documents is marked done; the one that remains, “All documents submitted to the IESG”, moved from December 2025 to 31 March 2027. The charter lets the group “optionally introduce the mechanism document at Experimental status first”.
The group held interims on 23 September and 21 October. The 21 October session page lists slides, a chat log and a recording but no minutes. Before it, the chair asked the list: “Are we converging on the WGLC being ready for the base document?” and “Do we believe that the keys/DNS document is also close?” Clayton answered on 19 October with five open topics: how spec-07 documents the verifier, whether to bring back donotmodify, whether Security Considerations need more text, whether DKIM2 needs a registry, and which signing algorithms to support and combine.
On 9 November, co-chair Murray Kucherawy wrote: “the chairs think we should aim to start WGLC on the “spec” document after we meet in San Francisco, though if there are still significant open issues we can of course defer that.” The IETF 127 session is on 18 November at 00:30 UTC (16:30 local on 17 November), a two-hour block. The agenda lists administrivia (5 minutes), the spec (Clayton, 20), CanModify (Clayton, 10), the DNS draft (Chuang, 10) and From field rewriting (Chuang, 20), then other business. On 10 November Clayton wrote that “the current version of the keys document is pretty final IMO though there has been limited discussion of how it impacts DKIM1”.
Who is working on DKIM2?
Participants write as individuals on a public list; affiliations are those shown on their messages.
- Bron Gondwana (Fastmail): lead motivator of the original drafts. On 19 November 2024 he wrote that “DKIM2 is designed to replace DKIM despite being a different protocol that does more than DKIM does. It’s supposed to entirely replace DKIM eventually.”
- Richard Clayton (Yahoo): lead author of the spec. The IETF 122 minutes record him saying: “It was not a design aim of DKIM2 to sweep up DMARC, ARC, greylisting, etc.; it’s meant to be a cleaner solution.”
- Wei Chuang (Google): lead author of the DNS and sender-policy drafts. Announcing an early DNS draft in July 2025, he wrote that “DKIM2 intends to be compatible with the existing DKIM installed base of keys” (quoted in a later reply).
- Allen Robinson (Google): in March 2025 he wrote that with two independent headers “DKIM1 verifiers would observe no change” (quoted in a reply). At IETF 126 he suggested requiring From alignment at the first hop or letting the originator signal it, and noted DMARC would need to recognize DKIM2.
- Todd Herr (GreenArrow Email): best-practices author; the draft says “Whether or not DMARC will be updated to include DKIM2 as an Authentication Mechanism is an open question at this time” (§7.2).
- Hannah Stern and Tobias Herkula (1&1 Mail & Media): press on encoding and
From:authenticity. - Vittorio Moccia: milter profile author; argued for two signatures over a composite.
- John Levine (Taughannock Networks): frequent reviewer; argued for multiple algorithms and a separate policy draft.
- Steve Atkins (Word to the Wise): Go implementer; raised DNS over TCP.
- R. Latimer (Inveigle.net): raised the
rt=privacy objection. - Murray Kucherawy and Pete Resnick: the chairs.
FAQ
Which DKIM2 drafts has the IETF adopted?
The specification, the DNS draft and the best-practices draft are working-group documents of the IETF DKIM working group. Other DKIM2 drafts are individual submissions with no formal standing until a working group adopts them. The status list on this page names each draft and its current revision.
What has to happen before DKIM2 becomes an RFC?
The working group first holds a Working Group Last Call, a final round of objections, and resolves what comes back. The documents then go to the Internet Engineering Steering Group (IESG), which approves IETF standards, and finally to the RFC Editor, which publishes them with an RFC number.
Who is working on DKIM2?
Participants write as individuals on the public ietf-dkim mailing list. Richard Clayton, Wei Chuang and Bron Gondwana author the core drafts, Todd Herr wrote the best-practices draft, and Murray Kucherawy and Pete Resnick chair the working group. Minutes and mail are public.
Changelog
Newest first.
- 14 November 2026. Moved here from the DKIM2 explainer. Added the WGLC plan, IETF 127 agenda, DNS-over-TCP, From-rewriting and DMARC-reports sections, the charter’s DMARC liaison line, the verifier’s envelope check (§14.4) and the note that the best-practices draft predates the current spec.
- 3 November 2026. Spec-09 added the Domain Policy Record and a NONE result for mail with no DKIM2 headers. The DNS draft was retitled, gained Clayton as co-author and dropped the composite key type.
- 1 November 2026. Added the ARC request’s second revision, which repeated the first with new dates. Spec-08 replaced the composite with plain ML-DSA-44 and reported a broken chain, a stale timestamp or a non-Authentic
From:as FAIL instead of PERMERROR. - 19 October 2026. Fourth sender-policy revision (dropped
unaligned, reversed §3.2) and spec-07 (removeddonotmodifyanddonotexplode). - 18 October 2026. DNS draft -01 (composite post-quantum key) and the missed ARC deadline.
- 1 October 2026. Fastmail’s signing announcement, the third sender-policy revision and the debug-header draft.
- 11 September 2026. Spec-06, the first sender-policy draft and the stalled ARC consensus call, then the Comcast, IIJ and Zone.ee replies to it.
- 4 May 2026. First published, as part of the DKIM2 explainer.
Corrections
- Result names. Earlier versions said a failed check returned “PERMFAIL”. The spec has no such state: it defines NONE, PASS, FAIL, PERMERROR and TEMPERROR (§14.1).
- Selector retention. An earlier version said the best-practices draft sets a number of days for keeping an old selector published. It does not.
- Latimer’s objection. An earlier version dated it 18 October 2025. The message is dated 14 October 2025 in New Zealand time, which is 13 October in UTC.
- A forecast that did not hold. The May version said to watch the DMARC list for the ARC Working Group Last Call, “likely in Q3 2026”. It did not come.