# The emerging role of AI in governance discussion > Source: > Published: 2026-08-26 04:50:50+00:00 A [discussion is underway on the IETF mailing lists](https://mailarchive.ietf.org/arch/msg/ietf/3VaBJ6pEdhtkpOtnZYA_HcVHejU/) about how to balance competing views on the use of AI in IETF work. Titled ‘Dealing with LLMs in IETF Discussions’, the [draft](https://datatracker.ietf.org/doc/draft-fengfar-led/) by Stephen Farrell and Chong Feng examines different perspectives on the inclusion of Large Language Model (LLM) generated text in IETF drafts and discussions. Using Internet drafts to stimulate discussion is a long-standing IETF tradition. Drafts do not bind the community to any particular course of action. They may remain as archived contributions, or they may attract enough interest to be taken up by a Working Group and eventually adopted. In that sense, they serve as a form of white paper, raising issues, concerns, ideas, and potential approaches for broader community consideration. At one level, this feels like an example of the ‘[boiling the frog](https://en.wikipedia.org/wiki/Boiling_frog)‘ phenomenon. Over the past 18 months to two years, the conversation has shifted from a strong presumption against the use of AI in this kind of discourse, to a softer view that its use may be impossible to prevent, and more recently to suggestions that, when used carefully, it may provide an overall net benefit. That is not to say the differences have disappeared. There remains a clear divide between those who want to exclude machine-generated text from drafts, standards work, and discussions, and those who see significant value in AI-assisted tools as a way to improve the clarity, comprehension, and expression of their ideas. Although unrelated to the IETF discussion, the EU is also considering a [Code of Practice on Transparency of AI-generated Content](https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content), reflecting broader concerns about how AI-generated or AI-modified works, including artwork, photographs, video, and text, should be identified in public media. The proposal sits within the framework of [Regulation (EU) 2024/1689, the EU Artificial Intelligence Act](https://ai-act-service-desk.ec.europa.eu/en/ai-act-explorer), which places transparency obligations on content providers and publishers in certain circumstances. The code of practice is intended to help organizations meet those obligations through the use of clear, widely understood labelling and disclosure mechanisms. Some of the ideas raised in this discussion echo similar approaches to disclosure and transparency. The specific mechanisms may differ, but the underlying intent is much the same. If authors disclose that they used an LLM to draft text, or relied on AI-based translation because they are concerned that using English as a second language may obscure their intended meaning, then the use of AI is made explicit. With that information available, readers can make an informed choice. They can read and engage with the contribution, or adhere to a principle of avoiding AI-assisted content. The latter may come at the cost of overlooking ideas that would otherwise contribute to their understanding of the proposal under discussion. Internet governance, as practised today through Regional Internet Registry (RIR) policy meetings, online policy discussion lists, proposals, and presentations, is conducted overwhelmingly in English. There is a certain irony in the term * lingua franca* being used to describe this state of affairs, given that the phrase itself originates from a Romance language. Yet the reality remains that English is the technical lingua franca for almost all of the work we do, providing a common means of communication across otherwise diverse language communities. When substantive policy is developed in other languages, such as Spanish or Portuguese in the LACNIC region, its wider exposure often depends on translation into English. At the same time, we know that a significant proportion of participants across the Internet community use English as a second language and routinely translate their ideas from their native language into this shared technical lingua franca. If AI tools can help these participants express their ideas more clearly and accurately in English, then we should be open to discussing and understanding that role. The goal should be effective communication of the underlying ideas, rather than creating unnecessary barriers for contributors whose first language is not English. It would be unfortunate if we convinced ourselves that AI and LLM-generated text played no role in the process, only to discover later that significant parts of proposals, analyses, or formal documents had been machine-generated. If such use is not disclosed, how would we know? IETF RFCs and standards are not laws in themselves, although many regulatory frameworks and binding requirements around the world reference, incorporate, or rely on them. What happens in standards development does not remain confined to the standards process. Over time, it flows outward into operational practice, regulatory expectations, and broader understandings of how systems should work. If the IETF ultimately accepts the use of LLMs and AI systems as a normal part of its standards development process, that acceptance will inevitably be reflected in the real-world use and interpretation of IETF documents. The norms established within the standards community rarely stay there; they tend to surface wherever those standards are implemented, referenced, or relied upon. So when the EU considers the growing role of AI in public discourse and calls for explicit disclosure of its use, it is at least conceivable that this expectation could extend to IETF documents that are considered for reference, incorporation, or use within telecommunications regulation. By the same token, the RIR policy process occupies a similar space. RIR policies routinely influence, and are influenced by, telecommunications regulation both internationally and within economies. If we are using these systems, it’s probably overdue we formalize how this is understood. The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a [Code of Conduct](/?p=395) applies to this blog.