IETF draft opens debate on LLM text in standards work

Stephen Farrell and Chong Feng have posted an Internet-Draft meant to start a broad conversation on how the IETF should treat large language model output in mailing-list traffic and documents. The draft is explicitly rough and non-prescriptive. It arrives as AI tools make it easy to produce plausible technical prose at volume, raising practical questions about accountability, originality, and whether readers can trust that a signed message reflects the sender's own judgment.

IETF executive director Jay Daley argued the draft underplays a central risk: when an AI surfaces ideas the human did not previously know, the author may trade genuine understanding for speed. He also challenged the claim that only humans open new conceptual territory while AI merely executes, noting that engineering mostly applies existing knowledge and that authors often mistake novelty to themselves for novelty to the field.

Chong Feng defended a refined version of the humans-originate line. The decisive step, he wrote, is not where a fact first entered someone's mind but who performs the abstraction, integration, and judgment that turn information into a technical claim the contributor can defend under review. Rob Wilton countered that frontier models have already produced novel mathematical proofs, so any hard boundary is already overstated.

Brian Carpenter held that tool use is irrelevant: the person whose name is on the message is fully responsible, as with a calculator or search engine. The real threat, he said, is floods of plausible Internet-Drafts and robotic mail that will eventually need technical countermeasures. Farrell strongly disagreed, saying readers need confidence that opinions are the sender's and not LLM mimicry, a burden not met merely by claiming responsibility. Daley sided with Farrell, observing that responsibility can devolve into simply piping every reply back to an AI for the next answer.

The draft has been revised to catalog arguments raised so far. Farrell suggested a dedicated list if no existing venue fits, but the exchange continues on the main IETF discussion list.