We are committed to respecting our users and contributors and maintaining the highest technical standards. This requires trust, transparency, and integrity.
This page outlines expectations for any contributions that involve the use of generative artificial intelligence ("GenAI") tools in order to uphold these values.
We recognize that the use of generative artificial intelligence technology (defined here as nondeterministic code, documentation, translation, and asset generators) is ubiquitous and controversial for many reasons. This policy document is to articulate our project's position on the submission of code, prose, and/or other types of contributions to our project that were created, in part or in full, using such tools. All other discussion (about these tools more broadly) is out of scope except for one point: fuck scrapers.
We believe the broader FOSS ecosystem will eventually converge on social and professional standards and expectations around how to responsibly integrate these tools with collaborative volunteer communities. While we hope this evolution does not cause serious harm to FOSS communities, we cannot predict or speculate on this matter. We know that each project will have different needs and requirements, whether in spirit or letter, and no policy will serve every project or community.
We do not attempt to address all possible edge cases. Our project does not take a position on issues not explicitly described in this document.
The purpose of this document is to set clear expectations to prevent or reduce confusion, friction, and legal exposure with respect to the use of these tools in producing submissions or contributions to our project.
The generative artificial intelligence landscape is evolving rapidly. We will update this document periodically to align it with our project's needs and to follow current best practices with respect to relevant legal opinions and industry standards. Please refer to version control history as needed.
This document assumes that it is impractical for us to reliably and accurately differentiate between human and machine output, that these tools will continue to become more useful and more difficult to detect, and that submissions made using these tools are inevitable.
The best we can do is to track source code provenance and include as much documentation as possible.
We also recognize that we cannot reasonably evaluate every contribution to every upstream project that we package (some upstream projects have embraced these tools; others have banned them wholesale. We treat canonical upstream projects as black boxes by default).
Our top priorities are to maintain the high standards of quality that our project has come to be known for and to maintain an environment of trust and transparency among contributors, users, and sister communities.
Our goal is to foster a collaborative environment where questions and curiosity are encouraged, clear and honest communication is expected, and all participants feel that their time and efforts are respected.
We believe transparency and integrity are key elements to a successful community project. This standard applies to maintainers as well as contributors. Transparency can foster learning, and integrity can create positive social forces.
We further recognize that shame and stigmatization are inconsistent with our project's values, disincentivize transparency, and do not promote constructive discussion or collaboration.
Our position is simple: do not claim credit for material that you did not author, do not misrepresent the origins of your contributions, and do not submit material you cannot fully explain or reproduce. If it feels like you're hiding something, err on the side of disclosure.
This disclosure policy is not new, but it is detailed more explicitly below.
We therefore discourage (but do not ban) the use of generative artificial intelligence tools, with one strict requirement: all contributions must be verified and submitted by a human.
By "verified" we mean the contributor must be able to explain every line of the contribution, demonstrate they wrote or authored it, and reproduce it from scratch if asked. Human oversight and accountability remain non-negotiable. This includes bug reports.
For code, we will not accept contributions that seem unnecessarily verbose, complex, or whose scope is larger than the problem at hand. Less is generally more; it is easier for us to review smaller changes and improves the likelihood of a successful contribution.
When contributing, please do not default to using these tools unless you are already familiar with our project and we are familiar with you. We want to avoid drive-by submissions of unknown provenance. This wastes our limited time, exposes us to unknown technical and legal issues, and reduces our ability to mentor new contributors.
Similarly, we'd prefer that a subject matter expert uses these tools over someone with less experience who is unable to verify the correctness of the output. Trust is earned.
One small carve-out for this strict policy is for cases involving language translation and accessibility. The same disclosure policy applies, however we are more likely to accept "assisted" translation contributions.
Our expectations with respect to contributing to our project using generative artificial intelligence tools are as follows.
We expect contributors to engage with our community in good faith.
Above all, we expect contributors to engage with our community in good faith. Regardless of how a submission was created, low-effort contributions and/or submissions that demand great effort to review create frustration for everyone. In general, current and potential contributors are expected to put in a reasonable effort to work with the community to find an agreeable path forward for the particular task at hand to avoid surprises.
We expect complete and thorough disclosure of the use of such tools when any generated output is to be submitted.
This includes but is not limited to: model(s), prompt(s), date(s) of access, inference provider(s) or software used (including version information), agent provider(s) or software used, software and session configuration (if applicable), session logs, etc. Date(s) of access are a non-negotiable requirement.
The requested information should be disclosed proactively so as to prevent doubts among community members who review your submissions. Since the purpose of these tools is to perform a task "better" and/or "faster" than a human, it is not unreasonable for us to ask you to track and disclose this information unless doing so would violate the terms of your inference provider. In such a case, we will not be able to accept your contribution. Please do not submit it, as doing so may increase legal exposure for both you and the project.
Please review our example disclosure template here: template.md
You remain responsible for your own submissions, and we don't want to host questionable material. Be sure to disclose enough context to allow maintainers or reviewers to assess potential technical and legal risks, verify provenance, and apply community standards.
Some coding tools automatically create git commits; this is an excellent way to contribute. We will squash down the final commits as needed. The purpose is to convey intent so that we are better able to help you help us. Again, please be sure that the final submission is submitted by a human.
Do not include any Co-authored-by: lines. It's a statistical model, and we don't credit compiler toolchains for "writing" machine code, either. We ask for the above information similar to how we would track down a compiler bug. Further, we don't wish to advertise for any company or tool, among many other reasons.
We expect contributors to adhere to our project's conventions, policies, and guidelines.
This includes abiding by all applicable licenses. If in doubt, ask our community for guidance and err on the side of caution.
Note that some tools are able to access third-party databases or search the Internet, and may indiscriminately plagiarize copyrighted or incompatibly-licensed material without your knowledge, independent of the data that the model was trained on. These licensing issues are serious and unresolved, but a paper trail can help. If you cannot provide the requested information, please do not submit your contribution.
We expect contributors to respect upstream project policies if patches are intended to be upstreamed.
If you intend to contribute to upstream projects, please ensure that any such contributions are also compatible with their respective policies. We will defer to the more restrictive policy. If an upstream project bans such contributions, we do not wish to be involved in laundering the code.
This is to maintain positive relationships with upstream projects.
We will not compromise on maintainability.
This is not a new policy. If your submission does not meet basic code quality and readability standards (regardless of technical correctness) it will be rejected. Low-effort contributions will receive low-effort rejections.
Many generative artificial intelligence tools produce output which can become unwieldy from a maintenance perspective. We do not have the time or interest to pick apart machine-generated submissions, since the machine will not learn from our advice in the same way that humans do.
We reserve the right to reject any submission for any reason.
This is not a new policy. We are grateful for your interest in contributing and we want to work with you, not against you, towards common goals. We want to respect your time; please respect ours.
If you are a new contributor and do not have a public reputation in other FOSS communities, we recommend spending some time engaging with our community before submitting large code or documentation changes.
Before spending time, money, or effort to contribute to our project using generative artificial intelligence tools, discuss your intentions with our community. We are thrilled to mentor new folks and encourage healthy brain development/maintenance.
We strongly encourage the use of any and all available tools (including generative artificial intelligence tools) to find bugs and other defects with the software that we distribute. One non-negotiable requirement is that you must verify all aspects of a bug report where such tools were used; this is intended to ensure that there is equity in the creation and review processes. This is because these tools raise many false positives.
Maintainers will review your submission with consideration for the nature and location of the change(s). For example, changes to a core cryptographic package may be evaluated under a different light than a game emulator with no reverse dependencies. Some maintainers may refuse to engage with "AI" contributions due to licensing concerns. Please check with the community (and respective maintainers) about your plan before attempting to use these tools.
Adélie is an independent Linux distribution committed to integrity, privacy, and user freedom. Your donation might be tax-deductible.