GlyphSigh. It's go time, folks: https://github.com/twisted/twisted/issues/12676
I've had a lot going on and have not wanted to have this difficult conversation with the community, but apparently we have reached the point where it absolutely needs to be laid out clearly. To be honest I am still kind of shocked that anyone would consider it acceptable to start putting LLM output into open source libraries at all, just purely based on IP concerns alone. The way that the vendors' IP indemnification works only makes sense for proprietary code.
This is definitely going to alienate some existing community members who don't agree with me on this, and further demoralize a community with precious few volunteer resources. So if you're the sort of person who'd like to see more projects adopt this stance and NOT get burned by a collapse in contributions as a result (and are not ALREADY spending half your life maintaining AI-free forks, you know who you are), consider helping draft policy language and jumping in to review PRs or fix bugs?
Octavia Con Amore 
Succubard's Library@glyph I'm glad more projects are banning LLM contributions 
banning LLM contributions will actually decrease the load on volunteers because there won't be an absolute deluge of junky contributions to review 
banning LLM contributions will actually decrease the load on volunteers because there won't be an absolute deluge of junky contributions to review 
Bytebro 🇬🇧 🇺🇦 🇬🇱I retired from work last year. I use Linux exclusively at home. I spent nearly half a century writing code in a zillion languages on umpteen operating systems.
Are there tips and tricks and/or how-tos, about how I might find a wee project that needs some TLC that I could perhaps work on and to which I might be able to usefully contribute?
I think I'd want to be 'upstream' if possible, so any benefits filter down to individual distros...
Just kinda thinking out loud here, sorry!
aburka 🫣
Christine Lemmer-Webber@glyph It's nice to not feel so alone, seeing another FOSS project lead say "we don't want our project to become a legal experiment"
I feel like the fact that we haven't seen enough leadership in FOSS say "hey, we DON'T know the licensing implications of this stuff" is catapulting the greater FOSS ecosystem into one giant legal experiment that seems, frankly, very likely to be grim for us.
I might say more in a separate thread. This has been brewing on my mind quite a bit lately.
Thanks for saying what you did in choosing to make a policy!
Peter Cock@cwebber @glyph another example https://github.com/biopython/biopython/pull/5241 (not agreed yet), similar rationale. A little dated but I blogged on this last year https://blastedbio.blogspot.com/2025/11/thoughts-on-generative-ai-contributions.html
eclexicI honestly don't know why so many projects think the legal repercussions will be nonexistent?
Probably for the same reason proprietary code projects are pretending they will have any legal standing to protect leaked code written primarily by LLMs...
I think this will get kicked down the road for a while, but I don't want to be involved when someone decides to pick it up...
The one thing is though, I'm unsure if copyright survives this in its current form.
@theeclecticdyslexic that could kick the legs out from today’s open source licensing… interesting times.
@theeclecticdyslexic @pjacock @cwebber so, I tend to agree the risk isn’t worth it, but I also understand why they think this. For one thing, the vendors have indemnification clauses (e.g.: https://cloud.google.com/blog/products/ai-machine-learning/protecting-customers-with-generative-ai-indemnification), so if you get sued for infringement you get some cover from your provider. They might think that this would apply if the project itself gets sued, although I am not so sure; in a proprietary case it gets more straightforward.
@theeclecticdyslexic @pjacock @cwebber There’s also a herd mentality which isn’t completely wrong, either. Open source license litigation is already extremely rare. The main people who *do* this litigation have openly advocated for accepting LLM contributions into your code. So who, exactly, is going to bring the lawsuit? https://sfconservancy.org/llm-gen-ai/llm-backed-generative-ai-recommendations.html
@theeclecticdyslexic @pjacock @cwebber that question does have an answer; it would be a new group of people. But they would have to form, and organize, and find an interested lawyer, etc. so it’s a big lift.
mirabilos🐈⬛@glyph @theeclecticdyslexic @pjacock @cwebber it also would still not make the result legal… the project would end then anyway
Ayush Agarwal (आयुष अग्रवाल)I honestly don't know why so many projects think the legal repercussions will be nonexistent?
Because in this hype cycle, corporations are the ones involved in flagrant violation of copyright and, believe it or not, enforcement of laws is the only thing holding some people back from doing the worst possible things you can imagine. I guess these people feel that even if there are negative legal consequences from using LLMs, corporations have bigger things to worry about rather than individuals, which is naive.
This is also why I feel that assigning any sort of importance to copyleft is pointless right now because it works within the confines of copyright and copyright itself exists in a flux these days. GNU and FSF are just wasting time by not adopting a strong anti-LLM policy.
I blocked an account on Fedi a few weeks ago which was asking people why they should respect robots.txt if they aren't legally required to do so. The QEMU people adopted LLMs with the following reasoning:
projects accepting AI-assisted content have not run into serious legal trouble so far, which suggests the probability of the risk materializing is not high
@mirabilos @ayushnix @pjacock @cwebber @glyph Ya, I don't think that anything that happens right now, related to the legal system, should be considered indicative of future performance... until such time as the bubble pops and geopolitics (and US politics) settles into something more stable (for better or worse).
Howard Chu @ Symas@glyph yep. Accepting AI-generated code into your projects is sheer idiocy. https://mastodon.social/@hyc/116596618413295067
SnoopJ@glyph congratudolences, welcome to the "I Thought This Exact Thing was the Entirety of Our Shit but I Guess I was Wrong" club
Piper Thunstrom
Janne Moren@glyph
Ultimately a project - a society - has to trust that people are honest. There is no effective way to detect dishonesty before the fact; we all rely on it coming to light often enough afterwards that it serves as a social deterrent.
Anybody who really wants to sneak in LLM code - or cheat on their taxes, or steal someone's lunch in the company kitchen - can do so. But most people don't want to be dishonest, and definitely don't want to be *outed* as dishonest to their peers.
@jannem I understand why nerds tend to immediately jump to this concern, but yeah, I don't share it for exactly this reason. I personally know and am even friends with tons of people (too many, really) who disagree with me and use LLMs all the time and think they're great. I hopefully know zero people who would lie to my face about using one; if I do, I don't want to be friends with them any more.
@glyph
And that's exactly the right approach. Most people *are* basically honest and *want* to do the right thing. Our entire societies are built on that foundation.
When you say "I don't want you to use LLM for this codebase" the vast majority of people - including LLM boosters - will say "OK, it's your code; that's fine with me." and respect that.
mx alex tax1a - 2020 (6)@atax1a @jannem unfortunately the soft booster in question is the second most prolific contributor to the project, which illustrates my previous issues. But more to the current point, he is completely honest about his usage, and advocating for a policy position which takes it into account, rather than lying to evade a policy.
Thankful Machine@glyph Good for you!! I am incredibly impressed with your ability to get out in front of the “how will you know” bullshit here:
https://github.com/twisted/twisted/issues/12676#issuecomment-4780248029
An example to be followed. Thank you.
Tobias
a very weeny construct 💀@glyph buncha others have said this already but i too think what you've written so far in that issue is very clear and well-reasoned
to such an extent that i would presume that anyone bringing vitriol in response must be arguing in bad faith. which might not be good diplomatic advice 🤷
@pho4cexa in this context I am not worried at all about the vitriol, the vitriolic responses I just block. It's the "oh well, I'm not going to waste my time writing code manually, guess I won't contribute any more" responses that I'm dreading
@glyph i want to say something crude and unhelpful about the kind of contributors that might disengage. but instead let's go with,
i hear you, and understand your worry, with empathy. i predict the project will be better off due to your thoughtful effort ❤️
Asko Nõmm
Lou.@glyph have you considered adding a .robots.txt file at the project root, following the same syntax as the one that would be found at the root of a webserver, with content saying bot are not to parse anything on any path?
Kind of the same idea as people using agents md files to say no to bots but using a convention that was made for telling no to bots.
Feral 3D@glyph I have a question. So LLM use on software development is already getting legally screwy and will be denagerousnand messy and dumb.
Why is there no FOSS llm projects that could be considered free and safe for use... legally at least.
Like public libraries should be making LLMs. It's our tax dollars anyway, should all LLMs be FOSS already !!?!
Linotype PilgrimCRITICAL SUPPORT 💚
synlogic4242@glyph thank you for taking a stand!
@synlogic4242 literally the least I could do