Yuan says the Claude attribution was "just a comment" and the code was
"written by hand." The emacs-devel list has accepted his word. But the
simple fact remains: a proprietary LLM's name is stamped on a commit in
Emacs master. It's now in the permanent git history. Yuan claims the
code itself is clean — but we have no way to verify that, and he's the
same person who put the LLM attribution there in the first place. His
statement is naturally tainted by his own action.
I am neither against Yuan neither the professional legal review of the
LLM policy in the GNU project. I am definitely for the project to follow
the policy, whatever it may be for group reasons. This e-mail however
discuss different issue.
Here I am talking about performative trust on emacs-devel mailing list.
Logic collapses on itself Eli trusts the same person who already broke
the emacs-devel rules to now self-certify that he only broke them in a
"harmless" way, and that is not any verification but face-saving
pretense. While commit speaks for itself, that it was co-authored by the
proprietary Claude Fable 5.
Everything else is just people telling each other what they want to
hear.
This email is for those paying attention:
- This is not about Yuan Fu personally. He was trusted with commit
access and that trust was sufficient. The issue is about the mailing
list itself (develipers) and how they handle trust.
- The developer Yuan appeared late in the discussion and offered a
simple declaration: the "Claude" attribution was a mistake, and the code
was written by hand.
- On the strength of that statement alone, the matter was declared
resolved. The mailing list has apparently settled on the position that a
remote developer — whose own commit message points squarely at LLM
contribution — must simply be believed when he says otherwise.
Fazit: Emacs-devel caught a developer with an LLM commit and resolved it
by pretending his word overrides his own commit message.
Yuan, thanks, it was entertaining.
Jean Louis
On 2026-07-25 17:18, Thanos Apollo wrote:
"Arsen Arsenović" <[email protected]> writes:
That some people, who don't know what the Luddites were, think that
calling someone a "Luddite" is an insult does not make them wrong.
Feel free to reclaim "Luddite" as a compliment, then. Good for you.
I'm not sure what you're trying to say here.
Obviously contributors are expected to respect policies? And those
with
write privileges trusted to respect them, even.
A policy that can distinguish only honest LLM users from dishonest ones
rewards nondisclosure while punishing honesty. It seems to be working
splendidly.
There is now a commit in master explicitly co-authored by Claude. If
that creates the legal problem you describe, the policy has already
failed.
Why not add an AGENTS.md defining an acceptable workflow, including
authorship and commit requirements? These tools are already widely
used. Refusing to document their acceptable use will not prevent it;
it
will merely encourage nondisclosure and make accidental violations more
likely.
How does that change whether or how the nonsensical property laws
apply
to LLM output?
It does not settle the law. But major software companies deploy
generated code at scale under well-funded legal departments. That is
strong evidence that the risk can be managed. "Some questions remain
unresolved" is not the same as "LLM-assisted contributions must be
presumed illegal."
--
Jean Louis
---
via emacs-tangents mailing list
(https://lists.gnu.org/mailman/listinfo/emacs-tangents)
