Hi again, Honestly, I think the friction here is that BCP 54 already covers much of what authors are expected to affirm implicitly. In particular, the expectation that “IETF participants use **their best engineering judgment** to find the best solution […]" excludes poor-faith submission of LLM synthesized output by construct, as LLMs are incapable of synthesizing genuine *judgement*.
If we are asked to look at someone's contribution by email, GitHub PR or someone's -00 or or even the -15 revision of someone's personal I-D, we are already expecting those submissions to be filtered by the authors' "best engineering judgement" in according to BCP 54. So a submission where "a substantial part of the document was created by AI" that risks misleading the viewer would already violate the relevant provisions in the existing Guidelines for Conduct, making the AI disclosure requirement redundant. Cheers, Nathanael On Fri, 5 Jun 2026 at 21:14, Brian E Carpenter <[email protected]> wrote: > On 06-Jun-26 14:06, John R Levine wrote: > >> Again, I'm not proposing this for copyright/legal reasons - that's > >> not RSWG business anyway - I'm talking about what RFC readers might > expeact > >> and care about. > > > > For about the fourth time, I think it is fine to ask, but if you're going > > to make it a requirement there's a lot more to do that I don't think we > > should do, at least not any time soon. > > > >> I decided to ask an expert, and here's what it said: > >> https://chatgpt.com/c/6a2379de-8228-83ec-8695-4c63d9fd08dd > > > > It said I had to log in. Nope. As I'm sure you know, it's tuned to be > > sycophantic and tell you that whatever you say is brilliant. > > Sorry about that - My question was "do people reading text want to know if > it was generated by a LLM?" > AI content is below my signature. > > Brian > > The evidence so far suggests a somewhat paradoxical answer: > > Yes, most people say they want to know. > Surveys consistently find large majorities favor disclosure when AI has > been used. For example, one 2026 international survey found 86% of > respondents wanted AI-generated content to be disclosed, and journalism > studies have reported similar numbers above 80%. [ > https://tribune.net.ph/2026/04/28/transparency-always-wins-a-new-global-study-finds-86-of-consumers-want-ai-generated-content-disclosed?utm_source=chatgpt.com > ] > > But disclosure often makes people trust the text less. > Several experiments have found that labeling content as "AI-generated" > reduces perceived accuracy, authenticity, and willingness to share it—even > when the content itself is unchanged, true, or even human-written. And > readers frequently cannot tell the difference on their own. > In blind tests, people often rate AI-generated content as equal to or > better than human-written content. Some studies found a preference for > AI-generated answers when the source was hidden, with the preference > declining once participants were told the content came from AI. [ > https://www.researchgate.net/publication/384533200_People_are_skeptical_of_headlines_labeled_as_AI-generated_even_if_true_or_human-made_because_they_assume_full_AI_automation?utm_source=chatgpt.com > ] > > So there are really two different questions: > > Do readers want transparency? > Generally yes. Many people view disclosure as an ethical issue and want > the information available. [ > https://arxiv.org/abs/2503.16458?utm_source=chatgpt.com] > > Does knowing improve their evaluation of the text? > Not necessarily. In many studies, the label changes perceptions more than > the content itself does. [ > https://www.researchgate.net/publication/384533200_People_are_skeptical_of_headlines_labeled_as_AI-generated_even_if_true_or_human-made_because_they_assume_full_AI_automation?utm_source=chatgpt.com > ] > > For a community such as the IETF, there is a further distinction. Readers > might not care whether an author used an LLM as a spelling checker, grammar > aid, or drafting assistant. They may care much more about whether the > technical content, protocol design, security analysis, and normative > requirements were actually reviewed and validated by the named authors. In > that setting, disclosure is often less about literary authorship and more > about accountability. > > In short: people generally want to know, but they do not necessarily use > that information rationally when judging the quality of the text. The > research literature currently points quite strongly in that direction. > https://www.researchgate.net/publication/384533200_People_are_skeptical_of_headlines_labeled_as_AI-generated_even_if_true_or_human-made_because_they_assume_full_AI_automation?utm_source=chatgpt.com > ] > -- > rswg mailing list -- [email protected] > To unsubscribe send an email to [email protected] >
-- rswg mailing list -- [email protected] To unsubscribe send an email to [email protected]
