W3C

– DRAFT –
Web & AI Interest Group

09 March 2026

Attendees

Present
AlexDawson, AndreiCiortea, AntoineZimmermann, cpn, DavideEynard, DavidWeekly, Dingwei, DoḡuAbaris, dom, EhsanT, fabien_gandon, gendler, janina1, JaninaSajka, JLemee, MihaiMaftei, PalakMathur, PierreAntoineChampin, PLH, Roy_Ruoxi, ShouqunLiu, tzviya, VagnerSantana, ZengLiang
Regrets
-
Chair
Fabien Gandon
Scribe
dom

Meeting minutes

GitHub issue notifications for the group

Fabien: we wanted to discuss whether to have github activity notified on the mailing list

Roy: right now, the public-webai mailing list receives a copy of any new issues and comments
… some people might feel this is too noisy
… any input on whether to change this to a weekly summary?

<dom> +1 to moving to a weekly summary

<tzviya> +1 to weekly

<Dingwei> +1

[support for moving to weekly]

Roy: I'll take care of doing

Andrei Ciortea on “Autonomous Agents on the Web”

Slideset: https://w3c-cg.github.io/webagents/Meetings/2026-03-09-WebAI-IG/WebAgents-CG.pdf

Andreai: I'm one of the co-chairs of the Web Agents CG, a professor at Switzerland University and a researcher at Inria
… [presenting other co-chairs]

<fabien_gandon> Andrei Ciortea: https://andreiciortea.ro/

[Slide 3]

Andrei: gather expertise from different areas: multi-agent, semantic web, etc

<fabien_gandon> Autonomous Agent CG: https://www.w3.org/community/webagents/

Andrei: goal is to inherit benefits of Web architecture, preserve beneficial properties of multi-agent systems, and be human centric

[Slide 4]

[Slide 5]

Andrei: revival of a longliving vision of agents on the Web

[Slide 6]

<fabien_gandon> W3C Web of Things (WoT): https://www.w3.org/WoT/

[Slide 7]

[Slide 8]

[Slide 9]

[Slide 10]

[Slide 11]

[Slide 12]

[Slide 13]

<Roy_Ruoxi> https://w3c-cg.github.io/webagents/TaskForces/Interoperability/Reports/report-interoperability.html

[Slide 14]

[Slide 15]

[Slide 16]

[Slide 17]

Dogu: are you going to focus on the standardization part for content creation for both agents and humans?
… we launched an org in that space recently

Andrei: at the moment, we're mostly surveying the area - looking at what exists, not developing new specs

Dogu: https://tg.community/

<fabien_gandon> Dogu shares: https://tg.community/

<Zakim> fabien_gandon, you wanted to ask for the targeted release date of the report

Fabien: in terms of the timeline for the report, when would you release a stable version of the report?

Andrei: hard to estimate - we've been slower than we hoped; we're building it as a living document with self-contained sections
… we might split the document into several reports to help
… we're working on the sections on identification
… we received a first contribution for environment interaction
… we're going to organize soon a session on accountability & policy - details will be announced to the mailing list

<Zakim> fabien_gandon, you wanted to ask about ethical aspects e.g. chapter 10 and 11

fabien_gandon: at the previous meeting, we had discussions with the Web Machine Learning WG work on the ethical principles they developed
… any discussion around ethics in the CG?

Andrei: one of the designal goal is agents alignment, which refers to the TAG work on Web user agents
… we plan to extract that into a document of its own

fabien_gandon: please feel free to share pointers to the IG who could also be a potential target to publish it more widely

Andrei: one last note: among the many other CGs working on AI Agents, our specific focus is on architectural design

<fabien_gandon> AI Agent Protocol Community Group https://www.w3.org/community/agentprotocol/

<fabien_gandon> Semantic Agent Communication Community Group https://www.w3.org/community/s-agent-comm/

AI Content Disclosure on the Web

Slideset: https://docs.google.com/presentation/d/1Q8zpjz1SWqxh3CYkHG7R9IdkJiETbe6ELjmvr-nnUoM/edit?slide=id.g289430bbe7c_0_173#slide=id.g289430bbe7c_0_173 (archived PDF copy)

David: I'm a technologist, used to be the W3C delegate at Capital One
… this is the first presentation of this proposal to a public audience
… I joined forces with Dogu who was working on a very similar proposal

Dogu: software engineer and student, teamed up with David to launch a CG on AI content disclosure

[Slide 1]

[Slide 2]

[Slide 3]

[Slide 4]

[Slide 5]

[Slide 6]

[Slide 7]

[Slide 8]

[Slide 9]

[Slide 10]

[Slide 11]

[Slide 12]

[Slide 13]

<Zakim> janina, you wanted to ask about real-time, unreviewed ai generated text

janina: useful categorization, but I have concerns with some of that given accessibility discussions we're having
… we're seeing a lot of AI-generated alt text generated on the fly - currently easy to spot because they're not really following WCAG guideline
… there are a number of situations where you don't want these automatically generated alt text - e.g. in educational context where accuracy of the alternative is critical

<Roy_Ruoxi> relevant issue: w3c/webai#5

janina: any thought to extend this to generated on the fly with no human review?

David: I think this would fit "autonomous" level - it could be applied to a <figcaption> element

<dom> [this wouldn't work to annotate the alt attribute though]

janina: likewise an automatically generated summary would be tagged as specifically autonomous?

David: indeed

<cpn> article ai-disclosure="none" is a bit odd, suggests there's no disclosure, rather than no AI used

Max: (co-chair hat off) general outline makes sense, I hope you'll consider creating a use cases/examples document that outlines the subtle differences between ai-assisted and ai-generated

<tzviya> +1 to gendler

Max: the distinction based on who started the authoring process can get tricky, and could better informed with specific examples to set expectations

David: you're right that's probably the fuzziest part of the proposal
… is it based on who did the first draft?

<Zakim> fabien_gandon, you wanted to ask for discussions about the incentives for adoption

Max: "1st draft" can take very different shapes - e.g. if I shared very rough notes as bullet points that is then fed to an AI for a written up memo... it would be great to get the sense of the CG on these

fabien: what would be the incentives to adopt these annotations? we've seen a number of great annotation formats that never took off for lack of adoption incentives

David: some of it is about trust, some of it about regulatory compliance
… if you lie, what are the consequences? what benefits might one get from volunteering these disclosures? e.g. boost on visibility
… I can imagine several carrots, and there is an array of regulatory bodies that want to enforce some of these that could act as sticks

<AlexDawson> Clarification for speaker: Green Claims are regulated in the EU. If you do lie about use of AI and AI does have greater impact than human implementation - you can face consequences.

fabien: PICS got better adoption than P3P because of these indexing incentives indeed

PLH: have you thought about applying this to the underlying code of the page? i.e. whether the UI of the page was partially or entirely generated by an AI

David: not thought about it at all; I expect in the upcoming few years, 100% of web content will be at least partially assisted by AI

<Roy_Ruoxi> Generative UI discussion

<fabien_gandon> +1 to the question of PLH and to code and content we can add datasets

David: in general, I suspect users don't care as much about how the page code was generated vs the level of accountability and review for the content itself
… code provenance is interesting though

dogu: the CG focuses on the content at the moment
… I've thought about using comments to annotate code

<gendler> The reader may not care about underlying code-creation, but perhaps infrastructure providers/pieces would care?

[from the chat: On the question about the level of AI engagement to be disclosed, there is a proposal on AI attribution that suggests a richer set of attributions: https://aiattribution.github.io/create-attribution ]

fabien: this could extend to the question of data provenance

Alex: had you thoughts about Schema which had mechanisms to mark advertisement disclosure per regulation?
… this could be re-used in the context of AI disclosure

david: the original proposal had a schema.org integration suggestion
… see dweekly/ai-content-disclosure

<fabien_gandon> Vagner Santana: On the question about the level of AI engagement to be disclosed, there is a proposal on AI attribution that suggests a richer set of attributions: https://aiattribution.github.io/create-attribution

Alex: did you consider annotating content coming from 3rd parties?

David: is this about annotating the target of the link?

Alex: situations where content is being imported

David: for links, it might be difficult to keep the annotation in sync between source and target - the browser could prefetch the info on link hovering

Alex: I'm coming at it from the sustainability perspective

pchampin: there may be a category missing: content not directly from humans nor from AI - e.g. content extracted from databases, which may be curated by humans. is this in scope?
… or data from a sensor?

David: it feels like this would be "autonomous" - because published without human review/accountability

<cpn> and now it's not an AI disclosure any more...

pac: my initial reaction was to distinguish deterministic vs generative

fabien: also plenty of room for discussion about what autonomy means for an agent

Fabien: We have opened a number of issues on the repo - please contribute thoughts and pointers

We are always looking for talks

Fabien: please get in touch if you have talks that would be interesting for the IG!

Max: next month meeting will be the first of the rotation cycle

<fabien_gandon> Thank you dom

<Roy_Ruoxi> s/@@@/https://github.com/w3c-cg/webagents/blob/main/Meetings/2026-03-09-WebAI-IG/WebAgents-CG.pdf/

Minutes manually created (not a transcript), formatted by scribe.perl version 235 (Thu Sep 26 22:53:03 2024 UTC).