This study seems to even suggest LLM performance is directly tied to plagiarism: https://www.sciencedirect.com/science/article/pii/S2949719123000213#b7 “We found that the models that consistently output the highest-quality text are also the ones that have the highest memorization rate”
Perhaps I’m reading all of these wrong, and I’m not a lawyer. This isn’t legal advice.
But this makes me worried about the future of KDE if LLM code isn’t disallowed for contributions. At least then it’d be the fault of people submitting it anyway, rather than KDE…
Sorry if this is the wrong place to bring it up or if this isn’t useful!
Ah, thank you! I think my focus would then be on this part, whether that works for LLMs:
Since many people, including e.g. the lawyer from my clip above, seem to believe this is essentially impossible (unless perhaps you have the entire training set to check).
Perhaps I may suggest that this should be revisited?
“You’re responsible for what you commit.” covers this case. If you, among “many people”, believes “this is essentially impossible”, then don’t. If someone else believes otherwise, then he probably could.
After all, without LLM, you can still copy other’s code into your merge request, and the KDE project won’t and can’t check. You are responsible.
So is the policy faith-based? I don’t quite understand. LLMs feel like something KDE should look at if they think it’s compatible or not, not the contributor.
It would be nice if the KDE project could have a full ban on GenAI contributions, for code as well as assets, artwork, writing, docs, and everything else. This technology is highly unethical for many reasons, from resource use, to societal effects (building big datacenters in communities disrupting power supplies or polluting the environment, for example), to climate effects, to being pushed by people with very unsavory agendas, and many other reasons.
All those reasons are valid reasons to reject GenAI, but I’ll mostly talk about the reason I personally care most about: that it is erasing the humanity of whatever it touches. GenAI takes in a massive amount of creative outputs of humans (code, art, writing, …) and soullessly regurgitates them, which I find highly offensive. The tech industry is performing a blitzkrieg on humanity itself.
I want a more human world, and this technology goes completely against that.
Another point that I feel should be made, is that this technology affects way more than software projects; it affects many professions, and in overwhelmingly negative ways, so I feel it is important that we ban it, in solidarity with writers, artists, journalists and many others.
I am aware that enforcement of full GenAI bans could be rather difficult, but that shouldn’t stop us from standing up against this technology. Enforcement could be done only on cases where there is no doubt that there is GenAI involvement. The important part, to me, is that the KDE project makes an ethical stance against it.
Free software already relies on a lot of goodwill, and there are already things we expect for which enforcement is even more difficult than enforcing GenAI bans. For example, if a contributor wrote code on company time, it could be that they do not own their contribution and therefore should not be contributing it.
From a liability standpoint it would make more sense to simply say “None allowed“ and place the burden of indemnification on the committer (since they violated your policy) and protects your plausible deniability rather than assuming any liability if a human reviewer affiliated with KDE fails to identify infringing code during the review process.
This feels like a disingenuous reply as my suggestion is very clearly scoped to AI-generated content within KDE and KDE maintainers - Not “All people“ and “All projects“ at “all costs“.
This is a complicated topic, but unenforceable policies aren’t useful.
A skillful user of LLMs produces code that’s 100% indistinguishable from hand-written code. You’ll never know. Thus, any blanket policy of “No LLM code” will incentivize them to lie, and nobody will be the wiser. I don’t think that’s a good policy.
An un-skillful LLM user will provide much more obviously “I’m an LLM” code and be unable to conceal it even if they try. But this isn’t currently a problem; KDE already gets merge requests like this and we reject most of them because the changes don’t work, the author can’t justify the architectural decisions behind the changes, and so on. So there’s no problem to be solved here; the system is working as expected to filter out bad-quality content.
As I have said in my earlier post, we already have policies that are much more difficult to enforce than GenAI bans, for example, we legally cannot accept code that was written on company time, or that for other reasons the author does not have the rights to. We rely on goodwill from contributors in this case. For GenAI, I feel it should be the same.
Yes, people can lie about usage of GenAI, and code (or artwork, assets, writing, etc) which includes (partially or fully) GenAI outputs will get merged. Very unfortunately, this cannot fully be prevented.
However, I feel this should not prevent us from making a stand against it on ethical grounds. (and there are many, many ethical reasons why one may be against this fundamentally abusive technology)
Even a policy that can’t be enforced very well is better than none. We are still making an ethical stand and we can ban obvious usage of the technology.
Fortunately, prolific users of this technology tend to be very vocal about it, so those will be easy to spot.
For those that try to sneak GenAI contributors in, I feel it is good that they have to lie about it. The harder it gets for them to use this technology, the better. They should not be able to be proud of using it, using it should be a shameful thing.
Where is this policy described? It’s fine if your company agrees or is sponsoring the work, obviously. In the case where you’re stealing time from your employer, this isn’t great, but precisely because it’s un-knowable and therefore unenforceable, to my knowledge we do not have a policy about it. Because it would be unenforceable. Because unenforceable policies aren’t a good idea.
If the point is making an ethical stand, then an unenforceable policy isn’t the right place for this. The right place would be in some kind of declaration of principles or voluntary community guidelines, such as the KDE manifesto or code of conduct.
I disagree. Policies that can’t be enforced breed disrespect for the rules.
Policies (and rules, and laws, and standards, and similar) that can’t be effectively enforced and exist more as general statements of principles about conduct should, IMO, be either changed so that they can be enforced, or re-stated in the form of principles, guidelines, or community standards. This way the consequence for breaking them isn’t some kind of formal punishment (which can’t be effectively meted out anyway) but rather intangible social stigma.
I’ve been leaning towards the idea that, hypothetically, if an LLM contribution is to be accepted, this outcome is exactly what should always happen, to the point that a policy could exist where any LLM contribution that doesn’t reach this outcome can be closed on sight.
In the history of the many projects that have adopted No-AI policies for contributors, this so far has not often happened. Additionally the same point could be made about the Developer Certificate of Origin (and similar clauses, like the one KDE has), which would be similarly unenforceable and would “incentivize lying” about licenses and code ownership. Because it is, effectively, a ban on third-party code of a different license, perhaps of proprietary codebases. And we would never be the wiser!
The system already works based on trust, I see no reason why it would work any less well here. People that are found guilty of not keeping to their promise about code ownership and those found guilty of not keeping promises around AI use should be treated as one and the same. Because under the hood, it’s a lot of the same questions about code ownership and licensing.
If we can justify it there, we can justify it here just for that reason alone. In that sense I don’t think this policy is “unenforcable”. If we can’t justify it here, then the clause about compatible licensing and code ownership might as well also be dropped because someone might lie.
I think this is a matter of wording. For example, QEMU has the following AI policy.
Current QEMU project policy is to DECLINE any contributions which are believed to include or derive from AI generated content. This includes ChatGPT, Claude, Copilot, Llama and similar tools.
A policy like this would be very much enforcable. Yes, some LLM contributions may pass (which is okay), but inherently it states that the project is against LLMs.
Isn’t this policy implicit from being a free software project? We wouldn’t want to accept code we cannot legally accept, no?
Fair, then the policy could be to ban everything that is clearly created (partially or fully) using GenAI. The manifesto and code of conduct could then state that the project is fully against all usage of GenAI.
The passion behind this proposal is understandable, but a blanket ban on all GenAI contributions to KDE would be misguided, counterproductive, and inconsistent with the values that make free software great.
The ethical concerns are real, but misdirected
Many of the harms cited — energy use, data center pollution, corporate power concentration — are legitimate criticisms of how GenAI is currently deployed by large corporations. But these are critiques of specific actors and infrastructure decisions, not of the technology itself. Open-source models run locally consume modest resources. A contributor using a small local LLM to help write documentation or spot a bug is not meaningfully responsible for Microsoft’s data center footprint. Banning the tool does nothing to hold the actual bad actors accountable.
“Erasing humanity” is a philosophical claim, not a demonstrated harm
The argument that GenAI “soullessly regurgitates” human creativity assumes that the process by which something is made determines its value, rather than the result. But KDE has never cared whether a contributor used an IDE with autocomplete, Stack Overflow, or copied a snippet from another FOSS project — as long as the contribution is good, appropriately licensed, and genuinely offered. A volunteer who uses an AI assistant to draft a first pass at documentation, then reviews, edits, and owns that output, has not “erased their humanity.” They’ve used a tool. That’s what humans do.
It sets a dangerous precedent for policing contributors
Free software thrives because it lowers the barrier to participation. A blanket GenAI ban would disproportionately disadvantage non-native English speakers who use AI to help write clearer documentation, neurodivergent contributors who use it to navigate unfamiliar codebases, or hobbyists who lack the time to do everything from scratch. Demanding that contributors prove their work is “purely human” is both unenforceable by the author’s own admission, and a form of gatekeeping that contradicts the inclusive spirit of open source.
The solidarity argument proves too much
The post argues KDE should ban GenAI in solidarity with writers, artists, and journalists. But KDE doesn’t ban proprietary software development tools out of solidarity with developers harmed by vendor lock-in, or refuse contributions written on Windows out of solidarity with users harmed by Microsoft’s practices. KDE’s mission is to produce excellent free software. Taking sweeping political stances on adjacent technology debates is beyond that mission and risks fracturing a community that contains people with a wide range of views.
A better path exists
The real concerns — licensing, attribution, code quality, and not laundering copyrighted material — can be addressed directly through existing mechanisms. Require contributors to certify their contributions are properly licensed. Hold all contributions to the same quality bar, regardless of how they were made. If a specific GenAI-generated contribution is low quality or legally dubious, reject it on those grounds. That’s principled and enforceable. A blanket ideological ban is neither.
KDE should trust its contributors, maintain its quality standards, and address genuine harms specifically — not chase a symbolic gesture that would make the project smaller, less accessible, and no more ethical.
