Park Suite 6, Hilton Park Vienna, Vienna, Austria.
Date/Time: 2026-07-23 16:30–18:30
Chairs: Ignacio Castro and Alvaro Retana
Notetakers: Laura Scalone, Ryo Yanagida, (add yourself)
Speaker: Chairs
10 min
Chairs called for notetaker; Laura and Ryo volunteered.
Speaker: Mark Nottingham, Cloudflare
25 min
Andrew Campling: Really interesting talk, on the sort of stick thing, I
think we're sharing a problem or experiencing a problem that the RARs
have had relatively recently but from a different angle, which is, I
think our rules are crafted on the assumption that everyone's engaging
in good faith.
... So basically, sort of upon your image we need a written branch
review of all the rules
Mark Notthingam: Agreed, more robust rules/policy and politics needs to
be considered
Dirk Kutscher: Thanks for the talk, just on the policy consideration. So
we in the IRTF can also do run experinments. So normally we do this for
technical specifications.
Mark: I'll propose experiments but cost consideration needed?
Dirk: about tooling; good start, and worth pursuing this direction. Best
practice/how to use it correctly is worth considering/exploring
Mark: I think the interesting thing is, a tool like this the LM could
make it a lot easier for policy makers to understand what we're doing.
Tommy Jensen: I came here after hearing about topic in plenary about
'right way to use it'. What do we about the talent-pipeline aspect of
the community? Can this be used to teach people?
Mark: I've started to do that, in terms of the dialogue that it
encourages you know it tries to actively discourage the just one and
done kind of you know throw this at the working group interaction.
Tommy: Bringing everyone up-to-speed with tools like this would be great
Mark: I think we probably need to be more explicit in the IETF about
that methodology as well. We probably need to do something like write
down what we mean by consensus
Brian Trammell: Adding to what tommy said; we should see the positives,
we should leverage the 'lowering barrier' aspect for the community
Mark: Thank you, I like how you described it in the terms of friction,
because that's how I think about this a lot and we want to reduce
friction for the things that we want to encourage and make sure that it
is accidentally reduced
Nick Doty: Following up on that, I agree with the problem statement. I
was confused with the rest of the talk. Effectivelly accelerating the
problem you just described. Tools can be used in a good way and bad way,
and might help with efficiency. We perhaps lose the opportunity to
engage with the community, if the machine can help people skip that
process. What are the negatives of the approach?
Mark: Nick, I hear you, and I've thought these thoughts as well. I'm not
a big AI booster, but it became violently clear to me that the work
groups that I've been participating in, people are already doing this a
lot, and i came to the realization that if theyre going to be doing that
and i cant stop them, we might as well give them tools that take them on
better paths.
Jay Daley: You enumerated bad behaviours, are you planning to publish
this? (??)
Mark: I think the IESG is interested in this topic and I'm happy to have
a chat with them. SO I think it's probably going to come up tomorrow
Jaime Jimenez: So I took notes just to remember so on the efficiency
tool so I mean is it an efficiency tool at heavy users, in my opinion,
quickly get saturated with the actual draft writing processes.
... And then the last one, my emails have sounded a bit AI-like I use it
a lot
Sarah Jennings: Q1. Have you seen how 'agreeable' the models are?
Perhaps the model shounds encouraging but community might be very
different Q2. How far back does the corpus go and is it useful as an
institutional memory store? And a comment: About the 'harrassment', use
of LLM is not exclusive to new comers, long time participants could very
much use it and still gets a freepass.
Mark: About the comment; yes, it's about how to use it. About the
question on 'is it agreeable?' I try and beat out of it. I'm sure I need
a bigger stick, especially with characterizing what's already happened.
Colin Perkins: Yeah the datatracker is useful back to about 2015. Fully
agree with all those comments, particular for new participants,
especially those for whom English is the native language means to be
very careful because this is probably a very useful tool for some of
them. We have an awful lot of guidelines RFCs if we hunt a lot of them
even provide useful guidance. It might be an interesting experiment to
do.
Carolina Caeiro: 3 points. Interested to learn IESG is working on this,
I have similar question to tommy about how to engage. Perhaps we don't
need to worry too much about new comers using this, this has a similar
tone to 'next gen are doomed' sort of narartive, perhaps it's more about
guidelines, where IESG can help. Third, about the process, we should be
on the side of humanity (?)
Mark: I think the issue is not so much about the newcomers themselves;
it's about institutional capacity. You know, we have the capacity to
process a certain number of standards, to have a certain number of
documents going through the RFC Editor queue
Speaker: Jaime Jiménez, Ericsson
15 min
Colin Perkins: Hi, Colin Perkins, this is very interesting. If I
understand correctly, you're putting all, a bunch of drafts from across
the IETF into this. What would be really interesting would be to look if
there are differences across the IETF in the typical rating state. Not
in terms of AI use,
Jaime: Agreed
Suncica Rosic: I'm a new comer, thank you for your presentation, I have
couple of question. Regarding the ensemble model use, I wonder better
you use majority voting or something else because in my experience some
models perform better on certain type of let's say sentences.
... Ok, but would it have been possible to create synthetic drafts based
on the ones before AI and then have parallel trends in place?
Jaime: (wrt comparison) that is what I'm doing. About Q1. I've compared
and seen some difference with sonnet vs Opus where Opus is more
rigorous. About Q2, on retrospect, I've done some statistical sampling
of the version 00 drafts pre-2022 as pre-AI. No perfect way to detec AI
so needs human-based feedback etc.
Jay Daley: Have you compared statistical analysis against any commercial
models? Like the public AI detection servies?
Jaime:
Sue Hares: You mentioned about how to do some pre-processing. (...)
Jaime: I've not done that
Chris Box: Saw your website. Are you paying for all the token fee? Is
this sustainable?
Jaime: I'm paying for it. It's just prototype. Do you think I can make
money? The cost is not crazy.
Priscylla Silva: Comment: I've been using these models to check,
usually, they are bad for papers many years ago. Have you tested them
with old RFCs?
Jaime: Not systematically, seemed to have lower pass rate. 6% and below.
Chairs: Are we interested in Interim?
Speaker: Marcelo Santos, Federal Institute of Sertão Pernambucano
25 min
Sue Hares: Thanks for the work. Comment about method: Based on the
manual analysis as well as statisticall analysis, one of the cause of
changes to RFC production is after a WG does WGLC, and sends it to IESG,
IESG process has many loops. The metholody is not capturing this well. I
recommend you check the methods, and examine why RFC production process
cause problem. You assume once WGLC passes, rest is constant, but that's
not true. Last point: WGLC has a problematic 'sub-cycle', I suggest to
look into it.
Marcelo:
Dirk Kutscher: I want to push back a little, in IRTF, we should not be
too fixated about RFC production. In the IRTF RFC production is not
necessary the goal.
Marcelo: Understood
Colin Perkins: Agree, this is interesting. I'd like to 'compare notes'
on how to find draft history. There are several inconsistent ways to
discover this and is challenging. Probably does not have a huge effect
but would like to compare with our methodology.
Marcelo: Sure
Jaime Jimenez: How do you see whether the work was happening beyond the
submission (??) (...) about the affiliation, it would be good to check
the affiliation and their country as well
Marcelo: Geographical information is hard. We took the TLD on ML emails.
I acknowledge that's noisy data.
John Levine: Interesting work. Drafts in IRTF necessary didn't fail when
it doesn't get published. This is true in IETF too — drafts could have
impact even if it didn't get published
Speaker: İlke İlhan, RIPE NCC
25 min
Peter Koch: I'm on those MLs, and in 90's generation. About remark about
'number alone is not enough, context is important' is a very good.
Usually for atlas, young researchers asking for help, old timers
responding. That's most of the traffic. In other working group, flood of
+1s. It definitely deserve further insights. Encourage to continue this
work
Ilke: Thank you
John levine: Interesting charts. Have you been able to track people with
email address changing? How many of the 'newcomers' just new email
address? In the IETF, we have datatracker that can help with this.
Another point; anecdotally, students might try joining but the others on
ML might be just running on autopilot and might not get the right
engagement. Existential issue is real but probably not solvable.
Colin Perkins: What would be interesting is to run sentiment analysis
tool and see if those bursts are people 'yelling' or actual discussions.
IETF certainly have MLs where people are 'yelling'.
Nick Doty: Thank you for the work. I'm curious about how people move to
another communication medium like github, slack etc. How do we track the
movement? Maybe in ML itself too? How should we look at different
medium.
Ilke: Good point. Worth considering.
Speaker: Colin Perkins, University of Glasgow
20 min
Draft: draft-perkins-analysing-sdo-data
Brian Trammell: Q1 yes, Q2 yes. Datatracker was built for operation but
defacto archive. You should perhaps raise it to IESG that archiving
matters.
Colin: will do
Priscylla Silva: Yes to two questions. Comment: it's hard to process all
these different data types/formats etc. Maybe my approach might help. I
read the draft last week, good start. When there are gap in the year,
this week I figured that people have multiple countries in DataTracker,
approx 4k people change countries during the meeting (!!). I started
writing a guide and notes in my project. I have some suggestions on
those issues. There are some AI/Data Science conference that has
check-list for pre-processing. Perhaps this is useful for people working
on IETF data. There are always people pointing out that there are issues
with data processing methods that comes from their own experience. Maybe
we need a checklist we can all use. e.g. most common misleading info
etc. maybe useful for the group?
Colin: Agreed
Jaime: People start by using API but rsync is better.
Colin: Some are only available via API. Extracting this incrementally is
really hard.
Chairs: we'll take this to the list about publishing this but first,
we'll discuss about adoptionon ML.
AI in Standards Participation
AI is increasingly used to research, write, and implement Internet
standards — and, increasingly, to participate in standards discourse
itself. As an assistive technology, it can lower barriers for those
facing language or other obstacles. However, it also makes it easy to
participate with limited human effort or understanding, straining a
community whose value rests on a finite pool of expert attention.
This talk demonstrates ietf-llm, a tool for querying a working group’s
or research group’s public record, and uses it to ground a discussion of
where AI helps participation, where it harms it, and what norms the IETF
and IRTF might adopt in response.
Mark Nottingham works on Internet and Web standards at Cloudflare. He is
a long-time contributor to the IETF and a member of the Internet
Architecture Board (IAB).
Analyzing AI Usage in IETF Drafts
Forthcoming.
Jaime Jiménez is a Master Researcher at Ericsson with over 15 years of
experience in IoT and Web technologies. He co-chairs the IETF CoRE
Working Group, contributing to the Constrained Application Protocol
(CoAP) and the device management protocol LwM2M. He is currently
developing autonomous generative AI agents, enhancing their reasoning
and integration with Web/IoT APIs and real-time data.
Related link: ietf.jaime.win/is-it-ai
From Draft to RFC: Mapping the Mortality of Internet-Drafts across the
IETF and IRTF
It is well known that most Internet-Drafts never become RFCs. This talk
shares early, work-in-progress findings from a study that looks at where
in the process drafts tend to stall, using the IETF Datatracker to
reconstruct each draft’s lifecycle, including its outcome, authors,
affiliations, geography, and process history.
So far, the results suggest that a draft’s fate is shaped largely at one
key point: Working Group adoption. Some patterns that appear clear in
simple comparisons, such as differences by region, by area, or between
the IETF and IRTF, seem to change once adoption is taken into account.
These are preliminary observations rather than firm conclusions, and the
analysis is still being refined. The aim of the talk is to share the
approach and initial signals, gather feedback, and discuss what they
might mean for lowering barriers to entry.
Marcelo Santos is a Professor and Researcher at the Federal Institute of
Sertão Pernambucano, Brazil. He holds a PhD in Computer Science from the
Federal University of Pernambuco, with part of his doctoral research
carried out at the IBISC Lab, Université d’Évry, France. His work
focuses on Internet measurement, community networks, innovation
ecosystems, IoT, and applied artificial intelligence. He has been
involved in initiatives related to digital inclusion, rural
connectivity, and technology-based innovation.
Community Vitality through RIPE Mailing Lists
RIPE mailing lists have been active for over 30 years and have been one
of the main communication channels for the RIPE community and the RIPE
NCC membership. This analysis looks into mailing list traffic over time,
the concentration of participation, the longevity of contributors, and
the role of the most popular mailing lists in overall discussion.
The results reflect long-term engagement within the RIPE community and
RIPE NCC membership, as seen through the mailing lists.
İlke İlhan is a Business Data Analyst at the RIPE NCC. Her work focuses
primarily on the RIPE NCC membership and the RIPE community.
Analysing Internet Standards Development Organisation Data
Draft
draft-perkins-analysing-sdo-data
This document outlines issues to consider when studying data relating to
the Internet standards development ecosystem. It identifies observable
components of standards development processes, proposes a taxonomy of
possible measurements, and highlights methodological, interpretive, and
ethical considerations.
The draft is intended to support a range of uses, including monitoring
standards development organisations, evaluating the evolution of
technical work, understanding technology deployment, and informing
community, leadership, and governance discussions.
Colin Perkins is a Professor at the University of Glasgow. His research
interests are in transport protocols, with a particular focus on
transport and signalling for real-time and interactive multimedia, and
on improving the way network protocols are designed, implemented,
specified, and standardised. He has been active in the IETF and IRTF
since 1996, has chaired the AVT, MMUSIC, and RMCAT working groups in the
IETF, and served as IRTF Chair from 2019 to 2025. He received his BEng
in Electronic Engineering in 1992 and his PhD in 1996, both from the
University of York.