Re: Dealing with LLMs in IETF discussions draft

On 8/8/2026 1:30 PM, Brian E Carpenter wrote:
> On 08-Aug-26 00:37, S Moonesamy wrote:
>> Hi Chong,
>>
>> One of the points in your email was about who makes the technical
>> claims in an Internet-Draft.  At the other end there is a person who
>> reviews the technical claims and dispute/disagree with those
>> claims.  That person (it can be more than one person) previously did
>> not use a LLM to assess the correctness of the claims or to submit
>> his/her disagreement.  Is it okay for the person to use a LLM to do 
>> that?
>
> I think that question is unimportant. What is essential is that the
> preson sending the message is 100% responsible for the message. Whether
> they used a slide rule, a pocket calculator, a dictionary, a spell 
> checker,
> a search engine, or an AI tool, to prepare the message is irrelevant.
>
>> IESG members sometimes send comments or questions as part of their
>> evaluation of an Internet-Draft.  Is it okay for an IESG member to
>> use a LLM to do that?
>
> Also unimportant. They are personally responsible for the result.
>
> In both cases, there is a remaining question - does the IETF want
> people to _acknowledge_ the use of AI in preparing messages or documents?
>
>> For information, the 2026 IEEE World Congress on Computational
>> Intelligence allows reviewers to use LLMs.
>
> Probably wise, since they couldn't stop them!

Once upon a time, it was not unheard off for supervisors to ask a 
subordinate to prepare a report, or portions of it, and then publish the 
report under their own name. It was also quite common for a professor on 
a program committee to ask a grad student to prepare a review of a 
paper. I don't whether similar use of subordinate labor happened for 
Internet drafts or RFCs. The common practice is to list all contributors 
as authors, or failing that to acknowledge them in the acknowledgement 
section. In any case, the same rule applies: if you publish something in 
your name, you own it. If the report that you said you produced gets 
your company in trouble, you, the supervisor, deal with the fallback. If 
an author protest against an incorrect review of their submission, you, 
the professor, will have to explain it to your peers. If your Internet 
draft makes ridiculous statements, it will affect the reputation of you, 
the nominal author, not that of your subordinate.

I don't see why the rules about LLM and other tools should be any different.

-- Christian Huitema