Summary

  • The IETF Secretariat created the General Area’s ai-in-standards non-Working-Group list on 9 September for discussion of AI use in the standards process. At the prewriting check, its public archive contained only that creation notice.
  • At IETF 126, the IETF Chair had said that a mailing list alone was insufficient and called for a facilitated structure supporting community input and decision making. The launch record names no facilitators, decision subjects, output states or authorized handoff. A public forum record could preserve those distinctions without pretending that discussion already equals policy.

Creating an address is a governance act

The Secretariat announcement is concise enough to read as plumbing. It supplies a list address, an archive, a subscription link, a purpose and an organizational home. The purpose is discussion of uses of artificial intelligence in the IETF standards process; the home is the General Area.

Those fields matter. A dedicated list gives a recurring dispute a stable public location. It prevents each conversation from having to discover its own venue. It lets an interested participant subscribe without following every topic on the general IETF list. It creates an archive whose chronology can later show what was proposed, challenged, revised or left unanswered.

The IETF’s mailing-list guide says that most of its work occurs on lists and that most are open to subscription, posting and public inspection. The non-Working-Group guidelines also place a small authority chain behind creation: an appropriate Area Director approves the request, the Secretariat maintains the list system, and list administrators act as moderators.

That is more than a chat room. It is still less than a decision mandate.

The launch notice does not call the list a Working Group or a Birds-of-a-Feather session. It does not say the list was created with an intent to form a Working Group, even though the general guidelines recognize that as one possible use of a non-WG list. It does not identify a chair, a consensus caller, an appeal route or an output. None of those omissions proves that the work is absent. They establish the limit of what the creation record currently authorizes a reader to infer.

July’s promise contained a second component

The analytical test comes from the IETF’s own IETF 126 plenary minutes. On 22 July, IETF Chair Roman Danyliw said AI was already changing how drafts are written and reviewed, how participants interact on lists and in meetings, and how consensus operates. He said the IESG had heard from participants, review directorates and Working Group chairs that the existing position was inadequate.

He then separated two jobs. Making a mailing list was the easy answer. The harder job was a facilitated structure that could gather community input and support community decision making about practices or norms. The possible output was deliberately open: an RFC, something more, or something less.

That statement should not be converted into a binding delivery commitment. It was an account of intended work, and it acknowledged uncertainty. But it supplies a public specification against which the next step can be read. The new list satisfies the address-and-archive component. Its one-sentence purpose does not yet satisfy the facilitation-and-disposition component.

This is not a semantic complaint about the word “forum.” A forum can host several different kinds of institutional work. Participants might exchange techniques for using AI to search an archive, propose norms for disclosure, design an experiment, debate moderation, write an individual Internet-Draft, ask GENDISPATCH for a process route or conclude that no common action is justified. Each outcome has a different owner and evidentiary status.

Without that state separation, activity can borrow authority from the venue. A widely repeated proposal may look like emerging policy. Silence may look like assent. A moderator’s operational choice may look like an IESG decision. A document posted to the list may look like a commissioned deliverable. None of those transformations is valid merely because the archive is official.

The first archive state is a baseline, not a verdict

At the cutoff for this article, the public ai-in-standards archive contained one message: the creation announcement. That observation is time-bounded. It does not mean the list has failed to attract interest, that messages are being withheld, or that the archive will remain quiet. It creates a clean starting state.

The date matters because the governance record can now grow from a known origin. A later reader should be able to distinguish the Secretariat’s setup message from participant contributions, facilitator summaries, consensus probes, formal decisions and external publications. Chronology alone will not make those roles legible. The archive needs typed events.

The IETF already has language for some of the surrounding boundaries. RFC 9245 explains why a dedicated list can take a topic that would otherwise live on the broad IETF discussion list. The new-work guide describes a non-WG list as one way to form a community of interest, alongside DISPATCH, a BoF, Area Director sponsorship and independent publication paths. Moving conversation into a dedicated venue does not select among those later routes.

The Note Well governs contributions, intellectual-property obligations and conduct. It attaches responsibilities to a message. It does not certify that its claims are true, that its sender used no AI, or that the institution endorses it.

Earlier discussion shows why facilitation cannot be decorative

The question did not arrive without evidence. A Research and Analysis of Standard-Setting Processes Research Group agenda described AI as both a tool that can lower language and other barriers and a means of participating with little human effort, placing pressure on finite expert attention. The session minutes record calls for more robust rules, possible experiments, explicit methodology and attention to institutional capacity.

That session was valuable input. It was not an IETF-wide decision. A research-group discussion, a plenary statement, a non-WG list and an eventual process document occupy different places in the authority chain. Combining them into one narrative of “the IETF decided” would erase precisely the governance work the new forum is supposed to perform.

The asymmetry around AI makes this especially important. A model can generate ten supportive messages at low marginal cost. Ten terse human objections might reflect deeper operational experience. The reverse is also possible: repeated alarm can crowd out evidence that assistance helps a participant write in a non-native language or retrieve institutional history. Counting posts cannot resolve either case.

RFC 7282 treats consensus as the work of understanding and addressing issues, not winning a head count. It also stresses that a consensus judgment needs reasoning and an identifiable caller. Its discussion of decisions on mailing lists does not mean that any mailing list can decide any subject. The mandate, scope and handoff still matter.

Give the forum a public state record

The minimum useful addition is not a long constitution. It is a small, versioned forum record next to the subscription and archive links.

First, it should state scope and exclusions. “AI in the standards process” can include drafting, review, retrieval, translation, code generation, participation, moderation, meeting operations and consensus assessment. If the forum will handle only some of these, participants should know where the rest belongs.

Second, it should name functions rather than confer vague status. Who administers the list? Who facilitates agenda formation? Who can summarize unresolved issues? Who, if anyone, can assess consensus, and for which output? Moderation authority over posts is not the same as authority to adopt a practice.

Third, it should identify possible output states. A proposal might be raised, evidence requested, experiment proposed, drafting, routed to GENDISPATCH, referred to the IESG, published independently, superseded, no action or closed with objections recorded. The exact vocabulary belongs to the IETF. The important feature is that progress cannot be inferred from message volume.

Fourth, evidence and objections need custody. An experiment should record its question, data boundary, error modes, correction route and decision relevance. A facilitator summary should link back to objections rather than replacing them. If an output advances elsewhere, the forum record should link the receiving process and stop claiming authority it no longer holds.

Finally, corrections should be visible. AI policy is likely to evolve faster than a static web page. A version history can show who changed scope, why a proposed norm moved state and which evidence caused a reversal.

This would not predetermine whether the IETF needs an RFC, a BCP, an IESG statement, a tooling experiment or no common rule. It would make those alternatives observable.

Openness needs a route, not just an entrance

RFC 3935 joins open participation to technical competence, quality and relevance. A dedicated list serves the first value by making the entrance visible. Facilitation serves the rest by turning contributions into issues that can be tested, answered and routed.

The distinction protects both newcomers and reviewers. A newcomer should not need a private sponsor to discover where a concern belongs. A reviewer should not be forced to treat every post as a request for the same level of institutional attention. A public state record can say that a contribution was received without saying it was accepted, say it was routed without saying it was endorsed, and say a proposal was closed without rewriting silence as consensus.

The 9 September launch is therefore real progress. The next evidence should not be a larger inbox. It should be the visible connector between discussion and disposition.

Sources

  1. IETF Secretariat — creation of the ai-in-standards list
  2. Current ai-in-standards public archive
  3. IETF 126 plenary minutes
  4. RASPRG agenda — AI in Standards Participation
  5. RASPRG session minutes
  6. IETF non-Working-Group list guidelines
  7. IETF mailing-list guide
  8. IETF guide to bringing new work
  9. RFC 9245 — IETF Discussion List Charter
  10. RFC 7282 — On Consensus and Humming in the IETF
  11. RFC 3935 — A Mission Statement for the IETF
  12. IETF Note Well
  13. Lu Heng — The Policy Mirror
  14. Lu Heng — Running Code Primary
  15. Lu Heng — Why BTW Media Exists