{"slug": "recommendations-when-using-llm-for-foss-contributions", "title": "Recommendations When Using LLM for FOSS Contributions", "summary": "Software Freedom Conservancy (SFC) issued urgent recommendations for Free and Open Source Software (FOSS) contributors who use Large Language Model generative AI (LLM-gen-AI) code assistants, urging them to prioritize license compliance and transparency. The recommendations, based on months of internal discussions and community consultation, aim to mitigate harms from proprietary AI systems while leveraging them to advance software freedom. SFC plans ongoing engagement including tutorials and Q&As to support FOSS projects navigating these tools.", "body_md": "The entire community of computer users, which quickly approaches\nevery human, faces the growing conundrum of generative artificial\nintelligence systems backed by Large Language Models (“LLM-gen-AI”)[ 1](#footnote-which-systems). Software freedom activists face\nparticularly difficult challenges in this regard; these LLM-gen-AI\nsystems have been applied in earnest to the endeavors of software\ncreation and modification.\n\nWe cannot sufficiently mitigate this tricky problem with merely one\nstatement or a few blog posts. In 2022, Software Freedom Conservancy\nbegan our journey on this particular issue when our policy fellow,\nBradley M. Kühn, published [If\nSoftware is My Copilot, Who Programmed My Software?](https://sfconservancy.org/blog/2022/feb/03/github-copilot-copyleft-gpl/). In the last\nyear, that journey grew in complexity and urgency when some of SFC’s\nmember projects and supporters began to regularly request moral and\nethical guidance on these matters. SFC spent these months in\nalmost-daily internal discussions about the plethora of dilemmas\npresented by LLM-gen-AI systems.\n\nIn 2024, [SFC\npublished an aspirational statement](https://sfconservancy.org/news/2024/oct/25/aspirational-on-llm-generative-ai-programming/), a thought experiment rather\nthan a definition. We now make urgent recommendations to those ordered by\ntheir employers to use LLM-gen-AI code assistants to contribute to Free\nand Open Source Software (“FOSS”).\n\nSome FOSS project leaders have taken a zero-tolerance approach to any LLM-gen-AI contributions to their projects. We support leaders who make such decisions. FOSS project leaders deserve our sympathy and understanding regarding the volumous onslaught of new contributions. Patch evaluation has always required careful analysis (after all, humans write bad code too). Now, that analysis demand (reasonably) feels daunting to maintainers. Everyone should respect their decisions.\n\nNevertheless, we cannot and must not ignore the many FOSS\ncontributors who decide to explore these tools for the betterment of\nFOSS. Software freedom activism only succeeds when we admit that we are\nat least decades away from universal software freedom. Proprietary\nsystems will continue to exist; there is a real danger they will\ncontinue to leapfrog FOSS. We should resist the use of proprietary\nsystems, which include the most popular LLM-gen-AI systems, but we\nshould also remain willing ([as\nwe always have](https://sfconservancy.org/blog/2026/jun/02/ethical-use-proprietary-develop-free-software-foss/)) to utilize such systems when they can advance\nsoftware freedom.\n\nAfter much study, consideration, collaboration, and consultation with many FOSS leaders, SFC formulated the following recommendations for FOSS contributors who have decided to use LLM-gen-AI systems to augment their FOSS work. We expect to update these recommendations periodically. These are not mandates, demands, conclusions, nor definitions; rather, they are best practices that we have formulated after careful study of the undeniable reality that some FOSS contributors do want to use these LLM-gen-AI systems.\n\nIn the months following the announcement of these recommendations, SFC plans an ongoing engagement campaign, including documents, online tutorials, public Q&As, and other community engagement, on these matters. SFC does not make these recommendations in isolation; rather, we offer sustained assistance to the community, particularly to FOSS projects working with proprietary LLM-gen-AI systems.\n\nThe long term goal of software freedom is to eliminate the harm of\nproprietary technology. While we work toward that greater goal, we\nshould seek to mitigate the harms that we cannot immediately eliminate.\nThese recommendations aim to abate the damage of these systems, and also\nconsider how these tools might counter-intuitively *help* us\nadvance FOSS.\n\nThese recommendations are listed in order of our view of their relative importance (most important first).\n\n**The FOSS community should support, not just tolerate,\nthose who outright reject LLM-gen-AI systems.** There are many\nintersecting ethical and moral issues regarding these systems, many of\nwhich are not currently fully understood. Anyone who chooses to avoid\nthem deserves our support and assistance.\n\n**Every FOSS contributor deserves self-determination\nregarding LLM-gen-AI.** No one should be *required* to use\nthese systems under duress. We make special note here of the increasing\nreports from technology workers who have been ordered by their\nmanagement (often under penalty of termination) to use these systems for\nall their work: FOSS and proprietary. Such mandates are unconscionable\nand we call on the industry to make use of LLM-gen-AI fully optional,\nand adopt non-discrimination policies regarding those who opt\nout.\n\n**FOSS projects should not shun contributors who choose to\nuse LLM-gen-AI systems.** Even FOSS projects that have chosen a\nzero-tolerance policy should make an effort to welcome contributors who\nsubmit a contribution that includes content or who received assistance\nfrom an LLM-gen-AI system. Such contributions should be treated no\ndifferently than a technically inadequate “first patch”: such submitters\nshould be welcomed to the community and receive a gentle (albeit perhaps\nform language) response thanking them for their interest and explaining\ngently why the project will not accept their contribution.\n\n**Before submission, FOSS Contributors must invest\nsubstantial time reviewing LLM-gen-AI -assisted and/or -generated\ncontributions.** Such contributions need curation. Contributors\nshould acquire an in-depth understanding of their contribution. FOSS\nprocesses yield software systems that are resilient, highly\nmaintainable, and contributor-friendly. Human contributors engage with\nFOSS projects (even as volunteers) because of the enjoyment and\nsatisfaction available in FOSS projects. **LLM-gen-AI\ncontributions could erode the best aspects of FOSS if an unsolicited\nonslaught of unvetted, prompt-generated contributions become\ncommonplace.**\n\n**Full disclosure of how and when an LLM-gen-AI system was\nutilized to assist in creation of a contribution is a moral\nimperative.** FOSS project leaders cannot make good decisions\nabout LLM-gen-AI policy if they cannot survey which contributions were\nassisted, and how much they are assisted. Part of the contribution\nprocess should (at least) include a disclosure of what LLM-gen-AI system\nwas used, its version (as these system change over time), and a brief\ndescription of how the system assisted the contributor. This information\nshould be included in a machine-readable format in commit logs.\n\n**Contributors should only submit “unattended”** FOSS maintainers are often\nvolunteers, or permitted to work only a limited amount of time on their\nupstream projects. Maintainers’ time is precious, and is best used in\nhuman-to-human interactions with new and existing human contributors.\nNew contributors should respect existing decisions about “unattended”\nLLM-gen-AI. Maintainers should think carefully about the types of\nunattended LLM-gen-AI contributions that may be useful. We encourage\nproject leaders to flexibly and regularly (but also slowly and\ndeliberately) consider policy changes on unattended contributions when\nnew contributors present new ideas.\n\n**LLM-gen-AI users should keep detailed and accurate\nrecords of their interaction and save those meta-artifacts for\nposterity.** LLM-gen-AI systems excel at automation of users’\nlogs of prompts, notes, and other written details of the interaction\nthat led to the creation of an artifact. FOSS contributors should keep\nsuch meta-artifacts, and *regardless of license* they should be\narchived as if they are part of the Corresponding Source for the\ncontribution. (In the coming weeks, SFC will publish tutorials and\ntemplates to assist in automating this important process.)\n\n**Avoid jumping to conclusions about the legal significance\nof generated contributions and whether they are\n“copyright-washing-machines that ruin copyleft”.** There remain\nmany unanswered legal questions, and experts are actively working on\nsolutions. SFC will publish more on this issue in the coming\nmonths.\n\n**Inputs impact the licensing of the artifacts**.\nThe question of licensing obligations for material passed through the\nprocess called “training” remains undecided. Nevertheless, most\nLLM-gen-AI sessions don’t begin *only* with a prompt. By\ncontrast, most commonly, the user points the LLM-gen-AI at a codebase\nand receives its assistance to generate a patch for that codebase. If\nthat codebase is under a copyleft license, your changes *must* be\nlicensed under the project’s license, due to both copyright and\ncontractual terms of that license.\n\n**“Copyleft Everything” remains the best viable and safest\napproach** Certainly those who want to release FOSS under\nnon-copyleft licenses have more to worry about when using these tools.\nIt’s apparent that every widely used LLM-gen-AI was trained on much\nwell-known copylefted code. Courts need years to deliver guidance on\nmany relevant legal questions. In the meantime, *nothing* stops\nyou from using a copyleft license for the work you generate,\nparticularly a license that is widely compatible with other copyleft\nlicenses. SFC will make its staff time available to the [copyleft-next](https://next.copyleft.org) project to eventually\noffer a license that is widely compatible with other copylefts and\nextremely suitable as a copyleft for LLM-gen-AI outputs.\n\n**When LLM-gen-AI systems (including proprietary ones) can\nmassively accelerate FOSS improvements, use of such tools is an\nappropriate strategic compromise**. Most FOSS developers are not\nexperts in the area of creation and training of LLM-gen-AI systems.\nThose developers should feel comfortable making the strategic choice to\nuse LLM-gen-AI systems in these cases.\n\nWe detest using proprietary tools and we are never comfortable\nrecommending them. Yet [for\nnearly fifty years, FOSS contributors have used proprietary tools](https://sfconservancy.org/blog/2026/jun/02/ethical-use-proprietary-develop-free-software-foss/) to\ncreate and advance software freedom. *Writing* proprietary\nsystems is undoubtedly an anti-social act that we all should avoid.\n*Using* proprietary systems, particularly when they can forward\nFOSS, is a highly fact-dependent tactical decision.\n\n**Warning**: do take great care to fully understand the\nimplications of any proprietary license. SFC will publish in the coming\nweeks some guidance on how to approach such analysis.\n\n**Those with skills and interest in making FOSS-friendlier\nLLM-gen-AI systems should do so as a matter of high priority**.\nWhile no system meets the (currently) [Impossible\nDream of our aspirational system](https://sfconservancy.org/news/2024/oct/25/aspirational-on-llm-generative-ai-programming/), there are obvious avenues of\npursuit that will make progress in that direction. SFC will highlight on\nour blog in the coming months individuals working in these\ndirections.\n\n**Do not overuse LLM-gen-AI, or allow your skills to\natrophy**. In our discussions with the FOSS community about\nLLM-gen-AI, there seems to be one universal conclusion: the systems are\nmost effective and help the most when a very experienced FOSS developer\nsits at the prompting helm. LLM-gen-AI systems should\n*complement* existing skills and tools, not *replace*\nthem. Developers should remain *curious* about why software acts\nthe way it does, and this curiosity should extend to the LLM-gen-AI\noutputs — and even the system itself.\n\n**Think carefully about your usage**. As software\ntechnologists, we have for decades made complex choices regarding\nresource consumption vs. convenience. The advent of CI, as just one\nexample, led to massive increases in computing time, while at the same\ntime simplified contribution workflows. As individual FOSS developers,\nwe are unlikely to change the bad behavior of these proprietary software\ncompanies who are either focusing on the creation of, or mandating the\nexcessive use of, LLM-gen-AI.\n\nThere are hundreds of intersectional issues of societal significance and social justice that are touched by these technologies, including the environmental impact of the development and use of these systems. Our focus and expertise centers on the implications for software; here we assess user freedom and add our ideas to the overall social conversations about how this technology should be used, controlled, and distributed in the context of FOSS. On matters unrelated to software freedom, we defer to experts that focus on environmental and other intersectional issues.\n\nIn our experience, FOSS contributors are historically much more mindful and concerned about how their actions impact others than developers of proprietary software and systems. Bring that mindfulness to your use of LLM-gen-AI. As just two examples:\n\ndon’t run to an LLM-gen-AI immediately for every problem,\n\npay attention to the LLM-gen-AI when it is clearly doing useless processing, and quickly redirect it to something more useful.\n\nMost new technologies have some adverse outcomes. We must carefully recognize and mitigate them. Social justice movements (including the software freedom and software right-to-repair movements) succeed when well-intentioned individuals act sustainably and consistently to bring needed change. In FOSS, those individuals constantly invent and improve new technologies that respect users’ rights and freedoms.\n\nThe recommendations above are a start. We’re ready for revision and further explanation as facts change. Our community has successfully deployed our unique acumen, and will again to shift this current imbalance of power. We must creatively act as we always have; the FOSS community excels at strategems that counteract proprietary software with ingenuity.\n\nSFC walks with you on this multi-generational journey to universal\nsoftware freedom and rights.\n\nExpect (but embrace) the trepidation as we take this next step together\nnow. SFC’s goal is steadfast: empower consumers and users to advance and\nexercise software freedom, and their right to software repair. Remember\nthese strategies have worked and **will continue to work**\nwhen we remain vigilant, mindful, and focused.\n\nSoftware Freedom Conservancy publishes this statement after months of internal deliberation and discussions with a group of volunteers, including John Sullivan, Stefano Zacchiroli, and many anonymous contributors. The statement was drafted by SFC in collaboration with that group.\n\n[ 1](#return-which-systems) These Recommendations are made specifically for systems (most of\nwhich are currently proprietary), such as Claude Code, Copilot CLI,\nAntigravity, and OpenCode. These systems utilize an LLM to generate\ntextual artifacts that assist software project contributors. There is no\nknown shorthand for referring to these systems, so we refer to them here\nas “LLM-gen-AI” to remind the reader throughout that these\nrecommendations are not necessarily intended to apply to other forms of\nAI nor other uses of LLMs\n\nBy “unattended”, we mean prompt-generated contributions that received\nno further human vetting.[↩](#return-unattended-means)", "url": "https://wpnews.pro/news/recommendations-when-using-llm-for-foss-contributions", "canonical_source": "https://sfconservancy.org/llm-gen-ai/llm-backed-generative-ai-recommendations.html", "published_at": "2026-06-18 16:14:56+00:00", "updated_at": "2026-06-18 16:31:29.863678+00:00", "lang": "en", "topics": ["large-language-models", "generative-ai", "ai-ethics", "ai-policy", "ai-tools"], "entities": ["Software Freedom Conservancy", "Bradley M. Kühn", "GitHub Copilot"], "alternates": {"html": "https://wpnews.pro/news/recommendations-when-using-llm-for-foss-contributions", "markdown": "https://wpnews.pro/news/recommendations-when-using-llm-for-foss-contributions.md", "text": "https://wpnews.pro/news/recommendations-when-using-llm-for-foss-contributions.txt", "jsonld": "https://wpnews.pro/news/recommendations-when-using-llm-for-foss-contributions.jsonld"}}