Publishing Community Group Plenary: Launching Digital Comics Task Force Minutes 15 October 2026 /4:00 PM JST | 9:00 AM CEST | 7:00 GMT | 3:00 AM EDT
Attendees
- Chair: Gautier Chomel
- Scribe: Wolfgang Schindler
- Attendance: 4 persons
Agenda
Publishers have brought forward the requirement from the EU to report and document the use of AI in the production of published materials. There are two Github issue on this topic: Reporting and documenting the use of AI in published documents and Identifying AI contributions to EPUB content
Notes
Gautier: Reporting and documenting the use of AI in publishing. Raised by a colleague from Taylor & Francis. New EU requirement telling us that the use of AI must be signalled to the user. With a specific concern about the vast quantity of AI generated alt texts in the context of remediation of ebooks to fit to other legal requirements. George transcribed that in an issue on the accessibility task force repo. Topic was also an issue in PMWG. A year ago this question was also raised by Wendy. In the next EPUB charter, we might raise the issue of EPUB metadata. As it might become a topic for PMWG, CG should list use cases and produce a note that PMWG might use.
Gautier: we’ve seen in issues comments several proposals to signal, that raises a second subject to consider: how do we make sure that we have a method to signal the use of AI in an interoperable way.
Brady: I made a comment at a Business Group meeting. A11Y data created by AI - I asked how to flag that. Brought it up with W3C people. We want to know for example about alt text generated by AI. It will be a lot of work to detect AI-generated information. That was a year ago, but it might have improved. I don’t know how valuable the labels are. I would like to get use cases for the WG before plunging into that work.
Ivan: This is an issue. A whole palette of shades of AI usage. In the WG, we are in a not in a position to create categories of AI use. We could discuss a hook for adding such information in EPUB. This is my personal view.
Gautier: We could make a review of what’s happening.
Ivan: It might be a topic for a PhD in a university.
Gautier: My understanding of the EU act is that it must be labelled as part of the content. We see it on music and video plateforms. In text, AI is helping the process in contrast to fully AI-generated content in music or video.
Brady: Does the EU requirement apply to websites? It feels like a Web problem. If it is a statement to be put in the content, that’s easy to do. If you hide it in metadata, it is not immediately visible to the user. If it is part of the metadata, it should be for the Web.
Gautier: in publishing, you have to display information about the book without having the book at hand - ONIX
Christopher: ONIX records allow to tell that it was created or assisted by AI. You may acquire some certification that a product is human-generated. In the EU AI Act beyond the contributor part, you also have an illustrator, reader, etc.
Ivan: Tools we develop for annotations might be of help here. Point to parts of the text where we might add special annotations about AI use. What does the use of AI imply - accepted, legal?
Brady: I do love using annotations for this. I’m afraid of going down this path. I recommend looking more closely at ONIX. We have people asking for better metadata. It’s not useful for great publishers, but for smaller companies.
Christopher: You may put bits of ONIX into EPUB already.
Gautier: Great idea, but external metadata is easier to update.
Ivan: Do we have to give some minimal instructions what a RS must do with these metadata? If it is just there, it might satisfy the legislation while breaking the spirit of the law.
Brady: I’m assuming that somebody in the legislation will tell us that we have to present it to the user, e.g. at the bookstore before a book is bought. The big players will take the external data. It’s a bit weird to have ONIX in two places - a live and a dead version that is not updated.
Gautier: Responsibility comes to the RS and the web site. It might be displayed in a country where the EU laws apply and might not be shown in another country where no legislation ask for it. Everyone on the Web will have this issue. When you raise the topic of annotation, you may have AI-generated annotations. It’s not only about AI but more generally about machine generated contents. My ACE-Report could be an annotation added to my EPUB. We all agree that we should not create new metadata, but we should document use cases for publishing and raise the attention for that issue.
Christopher: sounds good - use cases are very important
Gautier: I will open a template for usecase and publish it on the CG web page. Solutions should be interoperable with solutions outside of EPUB. Do you have any examples? One use case is alt text.
Brady: voice generation for audio books, video, translation.
Christopher: libraries have obligations in this issue (due to state funding), music industry, images with or without human review.
Brady: I wrote a text but have the style changed by AI afterwards.
Christopher: We get into nuances which are hard to define. We have to find out what people really need.
Gautier: in ONIX AI is considered as a contributor - thus as a named person. In music plateforms, you find a bunch of tracks under the artist name Miscellaneous, because the metadata says Miscellaneous is a contributor.
Christopher: You may actually be named.
Graham: an AI-generated voice, but some of these are actors that have sold their voice to AI labs. Some labs have a brand name for the voice, not for the LLM. It is not that useful for the book buyer to know the LLM used.
Gautier: Right, knowing which LLM is interesting for producers, but probably not much for customers. We will create one or two first use cases to start the collection. This collection is open to be enlarged further. Maybe do another meeting when we have sufficient material for discussion.
Action items
- Write a draft Explanatory document
- Write one or two use cases as starting point for the collection of use cases
Resume
- AI use covers a wide spectrum, from fully generated to AI-assisted content. The PCG’s role is to collect use cases in publishing and track related work so the community is aware, our use cases are considered, and we can inform which vocabularies to use in which context.
- Candidate use cases: AI-generated alt text, synthetic voice in audiobooks, translation, and images with or without human review.
- Signalling could appear in the content or metadata. Reading systems and websites would handle display, and legal obligations vary by country.
- ONIX can already flag AI involvement, and a chunk of ONIX could be included in an EPUB.