On Mon, 18 Aug 2025, Job Snijders wrote:
In view of the many ongoing lawsuits against AI companies by people who claim
that the training material for LLMs and in some cases the LLM output have
violated their copyrights, I don't see any way that you could make those
assurances for LLM generated material.
How about if you trained your own LLM on existing IETF RFCs only?
I'm not sure that resolves the potential copyright issues. The IETF's
policies are such that the author retains as many rights as possible.
If it's post 2004 RFCs with BCP 78 notices you'd probably be OK give or
take code with BSD licenses that require acknowledgment since our license
allows reuse in the IETF process. RFCs older than that are mostly a
copyright mystery. For some the Trust has licenses from the author or
employer, for most, they don't.
And/or IETF mailing list discussions? And/or some free software
available under permissive license? With a short acknowledgement a'la
"This draft contains text written by a LLM trained on all IETF RFCs."
(or similar) it seems that it could be argued to pass the above test.
Our mailing lists are IETF contributions, the rest, who knows?
Regards,
John Levine, johnl@xxxxxxxxx, Taughannock Networks, Trumansburg NY
Please consider the environment before reading this e-mail. https://jl.ly
