| [17:06:23] | <mstenta[m]> | > legal status of material produced by generative AI systems remains the subject of ongoing discussion in many jurisdictions, |
| [17:06:23] | <mstenta[m]> | as i understand it, this is one of the biggest potential issues for accepting LLM-generated code in GPL projects |
| [17:12:00] | <mstenta[m]> | someone summarized it really well in a comment i read recently... i wish i could find it but alas... |
| [17:14:18] | <mstenta[m]> | the basic point was: GPL (and copyleft licenses generally) rely on copyright in order for the license to be enforceable |
| [17:14:57] | <mstenta[m]> | and specifically, they rely on the fact that the code in GPL projects is copyrightable. if it isn't, then the rules of the GPL can't really be enforced |
| [17:15:30] | <mstenta[m]> | (requiring derivative code to be licensed under GPL too, etc) |
| [17:16:11] | <mstenta[m]> | so this isn't about "anti-AI" sentiment, or code quality concerns... it's a big legal question |
| [17:17:20] | <mstenta[m]> | and the easiest way GPL projects can ensure they don't end up in a legally untenable position, it seems, is to disallow AI-generated code entirely |
| [17:18:19] | <mstenta[m]> | relevant: |
| [17:18:23] | <mstenta[m]> | > The GNU Compiler Collection's refusal to accept AI-derived contributions is not, at its core, about code quality. It is about whether the GPLv3 license can survive at all when AI-generated code — which the US Copyright Office ruled in January 2025 is not eligible for copyright protection without meaningful human authorship — gets mixed into a project that uses copyright ownership to enforce its license terms. You cannot license what |
| [17:18:23] | <mstenta[m]> | you do not own. GCC's maintainers have decided they cannot afford to find out what happens when they try. |
| [17:18:28] | <mstenta[m]> | https://www.techtimes.com/articles/322384/20260731/copyright-void-gcc-ba... |
| [17:21:59] | <mstenta[m]> | > [The U.S. Copyright Office] concludes that the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements. This can include situations where a human-authored work is perceptible in an AI output or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts. The Office confirms that the use of AI to assist in the process of |
| [17:21:59] | <mstenta[m]> | creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability, and it finds that the case has not been made for changes to existing law to provide additional protection for AI-generated outputs. |
| [17:22:05] | <mstenta[m]> | https://blogs.loc.gov/copyright/2025/02/inside-the-copyright-offices-rep... |
| [17:23:12] | <mstenta[m]> | as far as i know, farmOS core has not accepted any PRs that were purely AI-generated yet... so this is important for us to consider |
| [17:25:47] | <mstenta[m]> | Greg: you have two open PRs that have not been merged yet which we may need to consider... https://github.com/farmOS/farmOS/pull/1079 and https://github.com/farmOS/farmOS/pull/1095 |
| [17:28:03] | <mstenta[m]> | More relevant bits from that Tech Times article: |
| [17:28:13] | <mstenta[m]> | > The GNU Project's framework for determining which contributions require formal copyright assignment to the Free Software Foundation uses a threshold of approximately 15 lines of code or text. Below that threshold, a patch is considered legally insignificant for copyright purposes — too brief to constitute an original creative work. Above it, the contributor must confirm they hold the copyright and have the right to license it to the |
| [17:28:13] | <mstenta[m]> | project under GPLv3. |
| [17:28:32] | <mstenta[m]> | > That threshold predates the AI era by decades. It was designed to manage a straightforward question: when does a patch need formal paperwork? But it maps onto the AI problem with uncomfortable precision. The US Copyright Office made clear that code generated predominantly by an AI, without meaningful human authorship, is not copyrightable by anyone — not the AI, not the company that made it, not the developer who prompted it. The |
| [17:28:32] | <mstenta[m]> | Supreme Court declined to disturb that ruling in March 2026. |
| [17:28:55] | <mstenta[m]> | > The consequence for a GPLv3 project is specific. GPL enforcement works because the copyright holder — in GCC's case, the FSF and the individual contributors — can take legal action against someone who copies the code into proprietary software without honoring the GPL's terms. If a portion of GCC is AI-derived and therefore has no copyright owner, that enforcement mechanism breaks down for those lines. |
| [17:29:10] | <mstenta[m]> | (I'll stop pasting chunks... read the full article...) |
| [17:35:14] | <mstenta[m]> | <mstenta[m]> "Greg: you have two open PRs that..." <- FWIW... #1095 is functionally under the 15 line "threshold"... it's the tests that push it over. And #1079 is largely a copy and paste and modification of code I already wrote... so.... 🤷😅 |
| [17:35:38] | <symbioquine[m]> | I haven't read the links you posted yet, but "not copyrightable by anyone" isn't always a disqualification - the same can be said for any trivial snippet of code. |
| [17:35:51] | <mstenta[m]> | but either way... this is something we need to consider moving forward... |
| [17:36:28] | <mstenta[m]> | symbioquine[m]: yea, sounds like GNU set a threshold of 15 lines of changes... "Below that threshold, a patch is considered legally insignificant for copyright purposes — too brief to constitute an original creative work." |
| [17:36:43] | <mstenta[m]> | "Above it, the contributor must confirm they hold the copyright and have the right to license it to the project under GPLv3." |
| [17:37:00] | <mstenta[m]> | this is all very fuzzy... |
| [17:37:12] | <mstenta[m]> | but i understand the need to reduce legal risk |
| [17:37:27] | <mstenta[m]> | also notably, "That threshold predates the AI era by decades." |
| [17:37:36] | <symbioquine[m]> | mstenta[m]: "Okay Claude, now refactor that change so the diff is less than 15 lines." 🫣 |
| [17:38:33] | <mstenta[m]> | sounds like a good topic for the next dev/monthly call(s) |
| [17:38:54] | <mstenta[m]> | thankfully, there are much bigger projects wading into these swamps ahead of us... |
| [17:39:46] | <mstenta[m]> | the Debian vote goes through Aug 28... curious to see where they land... but I would guess similar to GNU |
| [17:40:47] | <symbioquine[m]> | symbioquine[m]: "Hey Claude, implement a preprocessor lisp embedded DSL and rewrite the changes with that." |
| [17:40:48] | <mstenta[m]> | for anyone catching up, here's the tl;dr: the crux is the intersection of GPL (which requires copyright to work) and AI-generated code, which according to the US Copyright Office CANNOT be copyrighted |
| [17:44:12] | <mstenta[m]> | also notably, the Linux kernel arrived at a different decision: |
| [17:44:12] | <mstenta[m]> | > n April 2026, Torvalds and kernel maintainers established a policy that permits AI-assisted code, provided the contributing developer discloses the AI involvement, takes full responsibility for the output, and ensures the code meets the same quality standards as any human-written submission. |
| [17:44:50] | <mstenta[m]> | i wonder if this opens linux up to to legal risks... making it's own GPL license unenforceable |
| [17:45:52] | <mstenta[m]> | oh also interesting... GCC decided test don't matter... because they can be removed/rewritten if a copyright issue arrises |
| [17:46:19] | <mstenta[m]> | so they focus on "functional code" |
| [17:46:48] | <symbioquine[m]> | I doubt it makes it unenforceable for the entire codebase. Just risky to do something like a "full AI assisted rewrite in Rust". |
| [17:47:01] | <mstenta[m]> | true |
| [17:48:31] | <symbioquine[m]> | mstenta[m]: Annoying to have to tell contributors/consumers to be careful of what code they copy from the tests though... |
| [17:49:12] | <symbioquine[m]> | Can't assume that the whole codebase is attributable/reusable the same ways. 🫠😮💨 |
| [17:50:34] | <mstenta[m]> | yea, and ultimately small downstream projects (like farmOS modules) probably don't need to worry too much about any of this... because the likelihood of them needing to enforce their own license is small |
| [17:59:30] | <fosten[m]> | So very interesting. My parting comments at the end of the call on Friday was about the future of copyleft in the AI world, before I read any of this. Thanks for sharing! The philosophical questions intrigue me. |
| [18:05:48] | <symbioquine[m]> | Also interesting more broadly w.r.t. the idea of agentic coding is a way that "anyone can code". What kinds of code are worth writing if you don't end up with a copyrightable output...? |
| [18:18:27] | <mstenta[m]> | fosten: yea it is a very interesting collision, and throws copyleft as a concept into stark relief |
| [18:19:20] | <mstenta[m]> | tbh I've always been of the opinion that intellectual property is a flawed idea from the start ("owning" ideas), but the fact of the matter is GPL relies on IP law to function |
| [18:22:01] | <fosten[m]> | Last paragraph Debian thread “How will you enforce a ban on LLM contributions?"… In cases where it’s not obvious, how would you even know? |
| [18:23:52] | <mstenta[m]> | Yea a lot of this all relies on trust/faith that contributors are telling the truth... that has always been true |
| [18:24:09] | <symbioquine[m]> | You also can't necessarily know if code was plagiarized or copied from a proprietary codebase. |
| [18:24:55] | <symbioquine[m]> | * |
| [18:29:00] | <fosten[m]> | <symbioquine[m]> "Also interesting more broadly w..." <- Great perspective. If you can’t use the output then what’s the point if the utility is lost. |
| [18:29:45] | <symbioquine[m]> | fosten[m]: Actually I think it biases things towards useful code. |
| [18:30:13] | <symbioquine[m]> | Instead of coding to "create intellectual property". |
| [18:30:39] | <symbioquine[m]> | * things towards "useful code, * useful code". |
| [18:36:38] | <fosten[m]> | In learning AI you have these wow moments. Seems like this is one of those tipping points. Curious to see how it all trickles down. |
| [18:39:19] | <mstenta[m]> | Here's a question: if you use AI to write an app, can you legally sell that app? You don't own a copyright on it... |
| [18:41:05] | <mstenta[m]> | And if you build a business around a "proprietary" app that is AI generated... what is that business actually worth if you were hoping to "exit"? |
| [18:41:20] | <mstenta[m]> | The code can't be part of the valuation anymore, it would seem |
| [18:41:50] | <mstenta[m]> | (just wondering out loud) |
| [18:44:49] | <mstenta[m]> | And will GPL projects end up as the only remaining "clean" code that can be used for training data, if they all end up with no-AI policies? 😜 |
| [18:51:22] | <fosten[m]> | I wondered that too about commercial software. Two companies could easily have the same product. |
| [18:52:13] | <fosten[m]> | What if it’s 70% human and 30% AI. How about 60/40? Is that cool? Where’s the line. Would probably need a major lawsuit against Big Tech and their army of lawyers to determine that. |
| [18:53:46] | <mstenta[m]> | Yea all very fuzzy |
| [18:57:02] | <mstenta[m]> | Honestly (and setting aside the GPL questions for farmOS specifically) I kind of love that AI generated code can't be copyrighted |
| [18:57:10] | <mstenta[m]> | That is sort of remarkable in itself |
| [18:58:03] | <mstenta[m]> | Has there ever been such a huge amount of un-copyrightable works generated in such a short time before? |
| [18:58:36] | <mstenta[m]> | Interesting times! |
| [19:06:27] | <fosten[m]> | Also agreed. There are pros/cons. Takes time to sort through. And this is US copyright. Do other countries enact similar policies? Do businesses look for more favorable rules outside the US? |
| [19:07:03] | <mstenta[m]> | Good points and questions! |
| [19:17:10] | <mstenta[m]> | wow relevant: https://whoownsthecode.com/ |
| [19:42:16] | <symbioquine[m]> | <mstenta[m]> "Has there ever been such a..." <- Presumably the amount of animal generative output (also not copyrightable) is many orders of magnitude beyond that of anthropogenic works - though how much qualifies as "works" or "creative" is a matter for debate 🤓🍿🫣 |